A Machine Learning Study to Enhance Project Cost Forecasting

Abstract

In project management it is critical to obtain accurate cost forecasts using effective methods. This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we validate the model using three hundred experiments in the testing phase. Overall, the proposed model produces more accurate cost estimates when compared to the traditional Earned Value Management index-based model.

Full text

IFAC PapersOnLine 55-10 (2022) 3286–3291
ScienceDirect
Available online at www.sciencedirect.com
2405-8963 Copyright © 2022 The Authors. This is an open access article under the CC BY-NC-ND license
.
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2022.10.127
10.1016/j.ifacol.2022.10.127 2405-8963
Copyright ©
2022 The Authors. This is an open access article under the CC BY-NC-ND license
(
https://creativecommons.org/licenses/by-nc-nd/4.0/
)
A Machine Learning Study to Enhance Project Cost Forecasting
Tolga İnan*, Timur Narbaev**, Öncü Hazir ***
* Electrical-Electronics Engineering Department, Çankaya University, Ankara, Turkey (e-mail:
tolga.inan@cankaya.edu.tr)
** Business School, Kazakh-British Technical University, Almaty, Kazakhstan (e-mail:
t.narbaev@kbtu.kz)
*** Supply Chain Management and Information Systems Department, Rennes School of Business,
Rennes, France (e-mail: oncu.hazir@rennes-sb.com)
Abstract: In project management it is critical to obtain accurate cost forecasts using effective methods.
This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project
cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors
and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we
validate the model using three hundred experiments in the testing phase. Overall, the proposed model
produces more accurate cost estimates when compared to the traditional Earned Value Management index-
based model.
© 2022 IFAC (International Federation of Automatic Control). Hosting by Elsevier Ltd. All rights reserved.
Keywords: Cost forecasting, Earned Value Management, Estimate at Completion, Machine Learning,
Project Management.
1. INTRODUCTION
Almost all the projects experience cost overruns irrespective
of their size and type, as they face many uncertainties during
their life cycles. Various project monitoring and control
methodologies such as Earned Value Management (EVM) are
commonly used to limit these cost overruns. These methods
mainly support the organizations to monitor the progress of the
projects and budget use effectively. During the project
execution, at any time, project teams need to know about what
has happened since the project start and, more importantly, be
able to foresee what might happen in the remaining life of the
projects. This makes accurate estimates critical to completing
projects under budget and maintaining reliable communication
with project stakeholders.
However, project monitoring and forecasting decisions are
prone to the increasing uncertainties of today’s data-rich
business environments. In this respect, we observe the
considerable potential for using Artificial Intelligence
techniques in production (Cadavid et al., 2019; Rai et al.,
2021) and project control (Munir, 2019; Chen et al., 2020; Ong
and Uddin, 2020; Natarajan, 2022).
More specifically, Machine Learning (ML) algorithms can aid
organizations in enhancing project cost forecasting, which we
focus on in this study. Even though the potential benefits are
remarkable, the literature is still scant (Willems and
Vanhoucke, 2015; Pellerin and Perrier, 2019; Hashemi et al.,
2020). The following section will briefly introduce some
pertinent ML applications for cost forecasting in projects.
We first discuss the fundamentals of EVM, a Project
Management (PM) methodology used to measure and forecast
duration and cost in projects (Humphreys, 2018; Mahmoudi et
al., 2021). We note that the conventional index-based EVM
forecasting approaches assume linearity in cost growth
(Anbari, 2003; PMI, 2019). However, cost growth is usually
nonlinear in projects and often resembles an S-shaped curve
pattern (Barraza et al., 2004; Narbaev and De Marco, 2014b).
Moreover, the index-based methods may result in inaccurate
forecasts in the early stages of a project (Kim and Reinschmidt,
2011; Warburton and Cioffi, 2016) due to a few data points
available that make the extrapolation to the project’s
remaining part not reliable (Lipke et al., 2009; De Marco et al.,
2016). These two limitations of the index-based cost
forecasting methods motivate our study.
ML models have not been extensively applied for cost
forecasting in ongoing projects. However, ML has a great
potential to enhance decision making in PM (IPMA, 2020).
Organizations have been carrying out more and more projects
and gathering a tremendous amount of data from the
undertaken projects. In fact, along with the data, know-how is
also accumulated. Traditionally, this know-how is carried
from project to project by senior managers. Project managers
mainly depend on their experiences and implement traditional
methods to estimate the total cost and completion time and
take corrective actions using the predictions and relying on
their experiences.
Our study aims to demonstrate how managers can benefit more
from the data of the completed projects by using ML. In
particular, the projects' data with similar characteristics are
used to train ML-based estimators and support project
A Machine Learning Study to Enhance Project Cost Forecasting
Tolga İnan*, Timur Narbaev**, Öncü Hazir ***
* Electrical-Electronics Engineering Department, Çankaya University, Ankara, Turkey (e-mail:
tolga.inan@cankaya.edu.tr)
** Business School, Kazakh-British Technical University, Almaty, Kazakhstan (e-mail:
t.narbaev@kbtu.kz)
*** Supply Chain Management and Information Systems Department, Rennes School of Business,
Rennes, France (e-mail: oncu.hazir@rennes-sb.com)
Abstract: In project management it is critical to obtain accurate cost forecasts using effective methods.
This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project
cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors
and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we
validate the model using three hundred experiments in the testing phase. Overall, the proposed model
produces more accurate cost estimates when compared to the traditional Earned Value Management index-
based model.
© 2022 IFAC (International Federation of Automatic Control). Hosting by Elsevier Ltd. All rights reserved.
Keywords: Cost forecasting, Earned Value Management, Estimate at Completion, Machine Learning,
Project Management.
1. INTRODUCTION
Almost all the projects experience cost overruns irrespective
of their size and type, as they face many uncertainties during
their life cycles. Various project monitoring and control
methodologies such as Earned Value Management (EVM) are
commonly used to limit these cost overruns. These methods
mainly support the organizations to monitor the progress of the
projects and budget use effectively. During the project
execution, at any time, project teams need to know about what
has happened since the project start and, more importantly, be
able to foresee what might happen in the remaining life of the
projects. This makes accurate estimates critical to completing
projects under budget and maintaining reliable communication
with project stakeholders.
However, project monitoring and forecasting decisions are
prone to the increasing uncertainties of today’s data-rich
business environments. In this respect, we observe the
considerable potential for using Artificial Intelligence
techniques in production (Cadavid et al., 2019; Rai et al.,
2021) and project control (Munir, 2019; Chen et al., 2020; Ong
and Uddin, 2020; Natarajan, 2022).
More specifically, Machine Learning (ML) algorithms can aid
organizations in enhancing project cost forecasting, which we
focus on in this study. Even though the potential benefits are
remarkable, the literature is still scant (Willems and
Vanhoucke, 2015; Pellerin and Perrier, 2019; Hashemi et al.,
2020). The following section will briefly introduce some
pertinent ML applications for cost forecasting in projects.
We first discuss the fundamentals of EVM, a Project
Management (PM) methodology used to measure and forecast
duration and cost in projects (Humphreys, 2018; Mahmoudi et
al., 2021). We note that the conventional index-based EVM
forecasting approaches assume linearity in cost growth
(Anbari, 2003; PMI, 2019). However, cost growth is usually
nonlinear in projects and often resembles an S-shaped curve
pattern (Barraza et al., 2004; Narbaev and De Marco, 2014b).
Moreover, the index-based methods may result in inaccurate
forecasts in the early stages of a project (Kim and Reinschmidt,
2011; Warburton and Cioffi, 2016) due to a few data points
available that make the extrapolation to the project’s
remaining part not reliable (Lipke et al., 2009; De Marco et al.,
2016). These two limitations of the index-based cost
forecasting methods motivate our study.
ML models have not been extensively applied for cost
forecasting in ongoing projects. However, ML has a great
potential to enhance decision making in PM (IPMA, 2020).
Organizations have been carrying out more and more projects
and gathering a tremendous amount of data from the
undertaken projects. In fact, along with the data, know-how is
also accumulated. Traditionally, this know-how is carried
from project to project by senior managers. Project managers
mainly depend on their experiences and implement traditional
methods to estimate the total cost and completion time and
take corrective actions using the predictions and relying on
their experiences.
Our study aims to demonstrate how managers can benefit more
from the data of the completed projects by using ML. In
particular, the projects' data with similar characteristics are
used to train ML-based estimators and support project
A Machine Learning Study to Enhance Project Cost Forecasting
Tolga İnan*, Timur Narbaev**, Öncü Hazir ***
* Electrical-Electronics Engineering Department, Çankaya University, Ankara, Turkey (e-mail:
tolga.inan@cankaya.edu.tr)
** Business School, Kazakh-British Technical University, Almaty, Kazakhstan (e-mail:
t.narbaev@kbtu.kz)
*** Supply Chain Management and Information Systems Department, Rennes School of Business,
Rennes, France (e-mail: oncu.hazir@rennes-sb.com)
Abstract: In project management it is critical to obtain accurate cost forecasts using effective methods.
This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project
cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors
and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we
validate the model using three hundred experiments in the testing phase. Overall, the proposed model
produces more accurate cost estimates when compared to the traditional Earned Value Management index-
based model.
© 2022 IFAC (International Federation of Automatic Control). Hosting by Elsevier Ltd. All rights reserved.
Keywords: Cost forecasting, Earned Value Management, Estimate at Completion, Machine Learning,
Project Management.
1. INTRODUCTION
Almost all the projects experience cost overruns irrespective
of their size and type, as they face many uncertainties during
their life cycles. Various project monitoring and control
methodologies such as Earned Value Management (EVM) are
commonly used to limit these cost overruns. These methods
mainly support the organizations to monitor the progress of the
projects and budget use effectively. During the project
execution, at any time, project teams need to know about what
has happened since the project start and, more importantly, be
able to foresee what might happen in the remaining life of the
projects. This makes accurate estimates critical to completing
projects under budget and maintaining reliable communication
with project stakeholders.
However, project monitoring and forecasting decisions are
prone to the increasing uncertainties of today’s data-rich
business environments. In this respect, we observe the
considerable potential for using Artificial Intelligence
techniques in production (Cadavid et al., 2019; Rai et al.,
2021) and project control (Munir, 2019; Chen et al., 2020; Ong
and Uddin, 2020; Natarajan, 2022).
More specifically, Machine Learning (ML) algorithms can aid
organizations in enhancing project cost forecasting, which we
focus on in this study. Even though the potential benefits are
remarkable, the literature is still scant (Willems and
Vanhoucke, 2015; Pellerin and Perrier, 2019; Hashemi et al.,
2020). The following section will briefly introduce some
pertinent ML applications for cost forecasting in projects.
We first discuss the fundamentals of EVM, a Project
Management (PM) methodology used to measure and forecast
duration and cost in projects (Humphreys, 2018; Mahmoudi et
al., 2021). We note that the conventional index-based EVM
forecasting approaches assume linearity in cost growth
(Anbari, 2003; PMI, 2019). However, cost growth is usually
nonlinear in projects and often resembles an S-shaped curve
pattern (Barraza et al., 2004; Narbaev and De Marco, 2014b).
Moreover, the index-based methods may result in inaccurate
forecasts in the early stages of a project (Kim and Reinschmidt,
2011; Warburton and Cioffi, 2016) due to a few data points
available that make the extrapolation to the project’s
remaining part not reliable (Lipke et al., 2009; De Marco et al.,
2016). These two limitations of the index-based cost
forecasting methods motivate our study.
ML models have not been extensively applied for cost
forecasting in ongoing projects. However, ML has a great
potential to enhance decision making in PM (IPMA, 2020).
Organizations have been carrying out more and more projects
and gathering a tremendous amount of data from the
undertaken projects. In fact, along with the data, know-how is
also accumulated. Traditionally, this know-how is carried
from project to project by senior managers. Project managers
mainly depend on their experiences and implement traditional
methods to estimate the total cost and completion time and
take corrective actions using the predictions and relying on
their experiences.
Our study aims to demonstrate how managers can benefit more
from the data of the completed projects by using ML. In
particular, the projects' data with similar characteristics are
used to train ML-based estimators and support project
A Machine Learning Study to Enhance Project Cost Forecasting
Tolga İnan*, Timur Narbaev**, Öncü Hazir ***
* Electrical-Electronics Engineering Department, Çankaya University, Ankara, Turkey (e-mail:
tolga.inan@cankaya.edu.tr)
** Business School, Kazakh-British Technical University, Almaty, Kazakhstan (e-mail:
t.narbaev@kbtu.kz)
*** Supply Chain Management and Information Systems Department, Rennes School of Business,
Rennes, France (e-mail: oncu.hazir@rennes-sb.com)
Abstract: In project management it is critical to obtain accurate cost forecasts using effective methods.
This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project
cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors
and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we
validate the model using three hundred experiments in the testing phase. Overall, the proposed model
produces more accurate cost estimates when compared to the traditional Earned Value Management index-
based model.
© 2022 IFAC (International Federation of Automatic Control). Hosting by Elsevier Ltd. All rights reserved.
Keywords: Cost forecasting, Earned Value Management, Estimate at Completion, Machine Learning,
Project Management.
1. INTRODUCTION
Almost all the projects experience cost overruns irrespective
of their size and type, as they face many uncertainties during
their life cycles. Various project monitoring and control
methodologies such as Earned Value Management (EVM) are
commonly used to limit these cost overruns. These methods
mainly support the organizations to monitor the progress of the
projects and budget use effectively. During the project
execution, at any time, project teams need to know about what
has happened since the project start and, more importantly, be
able to foresee what might happen in the remaining life of the
projects. This makes accurate estimates critical to completing
projects under budget and maintaining reliable communication
with project stakeholders.
However, project monitoring and forecasting decisions are
prone to the increasing uncertainties of today’s data-rich
business environments. In this respect, we observe the
considerable potential for using Artificial Intelligence
techniques in production (Cadavid et al., 2019; Rai et al.,
2021) and project control (Munir, 2019; Chen et al., 2020; Ong
and Uddin, 2020; Natarajan, 2022).
More specifically, Machine Learning (ML) algorithms can aid
organizations in enhancing project cost forecasting, which we
focus on in this study. Even though the potential benefits are
remarkable, the literature is still scant (Willems and
Vanhoucke, 2015; Pellerin and Perrier, 2019; Hashemi et al.,
2020). The following section will briefly introduce some
pertinent ML applications for cost forecasting in projects.
We first discuss the fundamentals of EVM, a Project
Management (PM) methodology used to measure and forecast
duration and cost in projects (Humphreys, 2018; Mahmoudi et
al., 2021). We note that the conventional index-based EVM
forecasting approaches assume linearity in cost growth
(Anbari, 2003; PMI, 2019). However, cost growth is usually
nonlinear in projects and often resembles an S-shaped curve
pattern (Barraza et al., 2004; Narbaev and De Marco, 2014b).
Moreover, the index-based methods may result in inaccurate
forecasts in the early stages of a project (Kim and Reinschmidt,
2011; Warburton and Cioffi, 2016) due to a few data points
available that make the extrapolation to the project’s
remaining part not reliable (Lipke et al., 2009; De Marco et al.,
2016). These two limitations of the index-based cost
forecasting methods motivate our study.
ML models have not been extensively applied for cost
forecasting in ongoing projects. However, ML has a great
potential to enhance decision making in PM (IPMA, 2020).
Organizations have been carrying out more and more projects
and gathering a tremendous amount of data from the
undertaken projects. In fact, along with the data, know-how is
also accumulated. Traditionally, this know-how is carried
from project to project by senior managers. Project managers
mainly depend on their experiences and implement traditional
methods to estimate the total cost and completion time and
take corrective actions using the predictions and relying on
their experiences.
Our study aims to demonstrate how managers can benefit more
from the data of the completed projects by using ML. In
particular, the projects' data with similar characteristics are
used to train ML-based estimators and support project

 Tolgaİnanetal./IFACPapersOnLine55-10(2022)3286–3291 3287
Copyright ©
2022 The Authors. This is an open access article under the CC BY-NC-ND license
(
https://creativecommons.org/licenses/by-nc-nd/4.0/
)
A Machine Learning Study to Enhance Project Cost Forecasting
Tolga İnan*, Timur Narbaev**, Öncü Hazir ***
* Electrical-Electronics Engineering Department, Çankaya University, Ankara, Turkey (e-mail:
tolga.inan@cankaya.edu.tr)
** Business School, Kazakh-British Technical University, Almaty, Kazakhstan (e-mail:
t.narbaev@kbtu.kz)
*** Supply Chain Management and Information Systems Department, Rennes School of Business,
Rennes, France (e-mail: oncu.hazir@rennes-sb.com)
Abstract: In project management it is critical to obtain accurate cost forecasts using effective methods.
This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project
cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors
and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we
validate the model using three hundred experiments in the testing phase. Overall, the proposed model
produces more accurate cost estimates when compared to the traditional Earned Value Management index-
based model.
© 2022 IFAC (International Federation of Automatic Control). Hosting by Elsevier Ltd. All rights reserved.
Keywords: Cost forecasting, Earned Value Management, Estimate at Completion, Machine Learning,
Project Management.
1. INTRODUCTION
Almost all the projects experience cost overruns irrespective
of their size and type, as they face many uncertainties during
their life cycles. Various project monitoring and control
methodologies such as Earned Value Management (EVM) are
commonly used to limit these cost overruns. These methods
mainly support the organizations to monitor the progress of the
projects and budget use effectively. During the project
execution, at any time, project teams need to know about what
has happened since the project start and, more importantly, be
able to foresee what might happen in the remaining life of the
projects. This makes accurate estimates critical to completing
projects under budget and maintaining reliable communication
with project stakeholders.
However, project monitoring and forecasting decisions are
prone to the increasing uncertainties of today’s data-rich
business environments. In this respect, we observe the
considerable potential for using Artificial Intelligence
techniques in production (Cadavid et al., 2019; Rai et al.,
2021) and project control (Munir, 2019; Chen et al., 2020; Ong
and Uddin, 2020; Natarajan, 2022).
More specifically, Machine Learning (ML) algorithms can aid
organizations in enhancing project cost forecasting, which we
focus on in this study. Even though the potential benefits are
remarkable, the literature is still scant (Willems and
Vanhoucke, 2015; Pellerin and Perrier, 2019; Hashemi et al.,
2020). The following section will briefly introduce some
pertinent ML applications for cost forecasting in projects.
We first discuss the fundamentals of EVM, a Project
Management (PM) methodology used to measure and forecast
duration and cost in projects (Humphreys, 2018; Mahmoudi et
al., 2021). We note that the conventional index-based EVM
forecasting approaches assume linearity in cost growth
(Anbari, 2003; PMI, 2019). However, cost growth is usually
nonlinear in projects and often resembles an S-shaped curve
pattern (Barraza et al., 2004; Narbaev and De Marco, 2014b).
Moreover, the index-based methods may result in inaccurate
forecasts in the early stages of a project (Kim and Reinschmidt,
2011; Warburton and Cioffi, 2016) due to a few data points
available that make the extrapolation to the project’s
remaining part not reliable (Lipke et al., 2009; De Marco et al.,
2016). These two limitations of the index-based cost
forecasting methods motivate our study.
ML models have not been extensively applied for cost
forecasting in ongoing projects. However, ML has a great
potential to enhance decision making in PM (IPMA, 2020).
Organizations have been carrying out more and more projects
and gathering a tremendous amount of data from the
undertaken projects. In fact, along with the data, know-how is
also accumulated. Traditionally, this know-how is carried
from project to project by senior managers. Project managers
mainly depend on their experiences and implement traditional
methods to estimate the total cost and completion time and
take corrective actions using the predictions and relying on
their experiences.
Our study aims to demonstrate how managers can benefit more
from the data of the completed projects by using ML. In
particular, the projects' data with similar characteristics are
used to train ML-based estimators and support project
A Machine Learning Study to Enhance Project Cost Forecasting
Tolga İnan*, Timur Narbaev**, Öncü Hazir ***
* Electrical-Electronics Engineering Department, Çankaya University, Ankara, Turkey (e-mail:
tolga.inan@cankaya.edu.tr)
** Business School, Kazakh-British Technical University, Almaty, Kazakhstan (e-mail:
t.narbaev@kbtu.kz)
*** Supply Chain Management and Information Systems Department, Rennes School of Business,
Rennes, France (e-mail: oncu.hazir@rennes-sb.com)
Abstract: In project management it is critical to obtain accurate cost forecasts using effective methods.
This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project
cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors
and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we
validate the model using three hundred experiments in the testing phase. Overall, the proposed model
produces more accurate cost estimates when compared to the traditional Earned Value Management index-
based model.
© 2022 IFAC (International Federation of Automatic Control). Hosting by Elsevier Ltd. All rights reserved.
Keywords: Cost forecasting, Earned Value Management, Estimate at Completion, Machine Learning,
Project Management.
1. INTRODUCTION
Almost all the projects experience cost overruns irrespective
of their size and type, as they face many uncertainties during
their life cycles. Various project monitoring and control
methodologies such as Earned Value Management (EVM) are
commonly used to limit these cost overruns. These methods
mainly support the organizations to monitor the progress of the
projects and budget use effectively. During the project
execution, at any time, project teams need to know about what
has happened since the project start and, more importantly, be
able to foresee what might happen in the remaining life of the
projects. This makes accurate estimates critical to completing
projects under budget and maintaining reliable communication
with project stakeholders.
However, project monitoring and forecasting decisions are
prone to the increasing uncertainties of today’s data-rich
business environments. In this respect, we observe the
considerable potential for using Artificial Intelligence
techniques in production (Cadavid et al., 2019; Rai et al.,
2021) and project control (Munir, 2019; Chen et al., 2020; Ong
and Uddin, 2020; Natarajan, 2022).
More specifically, Machine Learning (ML) algorithms can aid
organizations in enhancing project cost forecasting, which we
focus on in this study. Even though the potential benefits are
remarkable, the literature is still scant (Willems and
Vanhoucke, 2015; Pellerin and Perrier, 2019; Hashemi et al.,
2020). The following section will briefly introduce some
pertinent ML applications for cost forecasting in projects.
We first discuss the fundamentals of EVM, a Project
Management (PM) methodology used to measure and forecast
duration and cost in projects (Humphreys, 2018; Mahmoudi et
al., 2021). We note that the conventional index-based EVM
forecasting approaches assume linearity in cost growth
(Anbari, 2003; PMI, 2019). However, cost growth is usually
nonlinear in projects and often resembles an S-shaped curve
pattern (Barraza et al., 2004; Narbaev and De Marco, 2014b).
Moreover, the index-based methods may result in inaccurate
forecasts in the early stages of a project (Kim and Reinschmidt,
2011; Warburton and Cioffi, 2016) due to a few data points
available that make the extrapolation to the project’s
remaining part not reliable (Lipke et al., 2009; De Marco et al.,
2016). These two limitations of the index-based cost
forecasting methods motivate our study.
ML models have not been extensively applied for cost
forecasting in ongoing projects. However, ML has a great
potential to enhance decision making in PM (IPMA, 2020).
Organizations have been carrying out more and more projects
and gathering a tremendous amount of data from the
undertaken projects. In fact, along with the data, know-how is
also accumulated. Traditionally, this know-how is carried
from project to project by senior managers. Project managers
mainly depend on their experiences and implement traditional
methods to estimate the total cost and completion time and
take corrective actions using the predictions and relying on
their experiences.
Our study aims to demonstrate how managers can benefit more
from the data of the completed projects by using ML. In
particular, the projects' data with similar characteristics are
used to train ML-based estimators and support project
A Machine Learning Study to Enhance Project Cost Forecasting
Tolga İnan*, Timur Narbaev**, Öncü Hazir ***
* Electrical-Electronics Engineering Department, Çankaya University, Ankara, Turkey (e-mail:
tolga.inan@cankaya.edu.tr)
** Business School, Kazakh-British Technical University, Almaty, Kazakhstan (e-mail:
t.narbaev@kbtu.kz)
*** Supply Chain Management and Information Systems Department, Rennes School of Business,
Rennes, France (e-mail: oncu.hazir@rennes-sb.com)
Abstract: In project management it is critical to obtain accurate cost forecasts using effective methods.
This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project
cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors
and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we
validate the model using three hundred experiments in the testing phase. Overall, the proposed model
produces more accurate cost estimates when compared to the traditional Earned Value Management index-
based model.
© 2022 IFAC (International Federation of Automatic Control). Hosting by Elsevier Ltd. All rights reserved.
Keywords: Cost forecasting, Earned Value Management, Estimate at Completion, Machine Learning,
Project Management.
1. INTRODUCTION
Almost all the projects experience cost overruns irrespective
of their size and type, as they face many uncertainties during
their life cycles. Various project monitoring and control
methodologies such as Earned Value Management (EVM) are
commonly used to limit these cost overruns. These methods
mainly support the organizations to monitor the progress of the
projects and budget use effectively. During the project
execution, at any time, project teams need to know about what
has happened since the project start and, more importantly, be
able to foresee what might happen in the remaining life of the
projects. This makes accurate estimates critical to completing
projects under budget and maintaining reliable communication
with project stakeholders.
However, project monitoring and forecasting decisions are
prone to the increasing uncertainties of today’s data-rich
business environments. In this respect, we observe the
considerable potential for using Artificial Intelligence
techniques in production (Cadavid et al., 2019; Rai et al.,
2021) and project control (Munir, 2019; Chen et al., 2020; Ong
and Uddin, 2020; Natarajan, 2022).
More specifically, Machine Learning (ML) algorithms can aid
organizations in enhancing project cost forecasting, which we
focus on in this study. Even though the potential benefits are
remarkable, the literature is still scant (Willems and
Vanhoucke, 2015; Pellerin and Perrier, 2019; Hashemi et al.,
2020). The following section will briefly introduce some
pertinent ML applications for cost forecasting in projects.
We first discuss the fundamentals of EVM, a Project
Management (PM) methodology used to measure and forecast
duration and cost in projects (Humphreys, 2018; Mahmoudi et
al., 2021). We note that the conventional index-based EVM
forecasting approaches assume linearity in cost growth
(Anbari, 2003; PMI, 2019). However, cost growth is usually
nonlinear in projects and often resembles an S-shaped curve
pattern (Barraza et al., 2004; Narbaev and De Marco, 2014b).
Moreover, the index-based methods may result in inaccurate
forecasts in the early stages of a project (Kim and Reinschmidt,
2011; Warburton and Cioffi, 2016) due to a few data points
available that make the extrapolation to the project’s
remaining part not reliable (Lipke et al., 2009; De Marco et al.,
2016). These two limitations of the index-based cost
forecasting methods motivate our study.
ML models have not been extensively applied for cost
forecasting in ongoing projects. However, ML has a great
potential to enhance decision making in PM (IPMA, 2020).
Organizations have been carrying out more and more projects
and gathering a tremendous amount of data from the
undertaken projects. In fact, along with the data, know-how is
also accumulated. Traditionally, this know-how is carried
from project to project by senior managers. Project managers
mainly depend on their experiences and implement traditional
methods to estimate the total cost and completion time and
take corrective actions using the predictions and relying on
their experiences.
Our study aims to demonstrate how managers can benefit more
from the data of the completed projects by using ML. In
particular, the projects' data with similar characteristics are
used to train ML-based estimators and support project
A Machine Learning Study to Enhance Project Cost Forecasting
Tolga İnan*, Timur Narbaev**, Öncü Hazir ***
* Electrical-Electronics Engineering Department, Çankaya University, Ankara, Turkey (e-mail:
tolga.inan@cankaya.edu.tr)
** Business School, Kazakh-British Technical University, Almaty, Kazakhstan (e-mail:
t.narbaev@kbtu.kz)
*** Supply Chain Management and Information Systems Department, Rennes School of Business,
Rennes, France (e-mail: oncu.hazir@rennes-sb.com)
Abstract: In project management it is critical to obtain accurate cost forecasts using effective methods.
This study presents a Machine Learning model based on Long-Short Term Memory to forecast the project
cost. The model uses the seven-dimensional feature vector, including schedule and cost performance factors
and their moving averages as a predictor. Based on the cost variation patterns from the training phase, we
validate the model using three hundred experiments in the testing phase. Overall, the proposed model
produces more accurate cost estimates when compared to the traditional Earned Value Management index-
based model.
© 2022 IFAC (International Federation of Automatic Control). Hosting by Elsevier Ltd. All rights reserved.
Keywords: Cost forecasting, Earned Value Management, Estimate at Completion, Machine Learning,
Project Management.
1. INTRODUCTION
Almost all the projects experience cost overruns irrespective
of their size and type, as they face many uncertainties during
their life cycles. Various project monitoring and control
methodologies such as Earned Value Management (EVM) are
commonly used to limit these cost overruns. These methods
mainly support the organizations to monitor the progress of the
projects and budget use effectively. During the project
execution, at any time, project teams need to know about what
has happened since the project start and, more importantly, be
able to foresee what might happen in the remaining life of the
projects. This makes accurate estimates critical to completing
projects under budget and maintaining reliable communication
with project stakeholders.
However, project monitoring and forecasting decisions are
prone to the increasing uncertainties of today’s data-rich
business environments. In this respect, we observe the
considerable potential for using Artificial Intelligence
techniques in production (Cadavid et al., 2019; Rai et al.,
2021) and project control (Munir, 2019; Chen et al., 2020; Ong
and Uddin, 2020; Natarajan, 2022).
More specifically, Machine Learning (ML) algorithms can aid
organizations in enhancing project cost forecasting, which we
focus on in this study. Even though the potential benefits are
remarkable, the literature is still scant (Willems and
Vanhoucke, 2015; Pellerin and Perrier, 2019; Hashemi et al.,
2020). The following section will briefly introduce some
pertinent ML applications for cost forecasting in projects.
We first discuss the fundamentals of EVM, a Project
Management (PM) methodology used to measure and forecast
duration and cost in projects (Humphreys, 2018; Mahmoudi et
al., 2021). We note that the conventional index-based EVM
forecasting approaches assume linearity in cost growth
(Anbari, 2003; PMI, 2019). However, cost growth is usually
nonlinear in projects and often resembles an S-shaped curve
pattern (Barraza et al., 2004; Narbaev and De Marco, 2014b).
Moreover, the index-based methods may result in inaccurate
forecasts in the early stages of a project (Kim and Reinschmidt,
2011; Warburton and Cioffi, 2016) due to a few data points
available that make the extrapolation to the project’s
remaining part not reliable (Lipke et al., 2009; De Marco et al.,
2016). These two limitations of the index-based cost
forecasting methods motivate our study.
ML models have not been extensively applied for cost
forecasting in ongoing projects. However, ML has a great
potential to enhance decision making in PM (IPMA, 2020).
Organizations have been carrying out more and more projects
and gathering a tremendous amount of data from the
undertaken projects. In fact, along with the data, know-how is
also accumulated. Traditionally, this know-how is carried
from project to project by senior managers. Project managers
mainly depend on their experiences and implement traditional
methods to estimate the total cost and completion time and
take corrective actions using the predictions and relying on
their experiences.
Our study aims to demonstrate how managers can benefit more
from the data of the completed projects by using ML. In
particular, the projects' data with similar characteristics are
used to train ML-based estimators and support project
managers in making estimations. These estimation methods
can improve forecasting practices by considering the
nonlinearity in cost growth and making estimates by studying
the given data points. To show the effectiveness of the chosen
approach, we will compare the accuracy of the cost forecasting
results of our ML approach with the ones by the traditional
index-based model.
The remainder of the paper is structured as follows. Next, we
introduce key EVM metrics and review relevant studies on ML
applications in project cost forecasting. Then, we present our
ML approach and the project dataset for calculating the cost
estimates. Next, we report the results of our comparative
analysis and discuss the main results. Finally, we conclude
with the study summary, research limitations, and future
research directions.
2. BACKGROUND
2.1 Key EVM metrics
According to the Project Management Institute (PMI, 2019),
EVM is a methodology used by project managers to monitor
and control the schedule and budget of a project. It is based on
three key metrics: Planned Value (PV) – the budgeted value of
the scheduled work; Actual Cost – the actual value of the
performed work; and Earned Value (EV) – the budgeted value
of the performed work (Anbari, 2003). Budget at Completion
(BAC) is the project’s total budget and Cost at Completion
(CAC) is its total actual cost at completion. Planned Duration
(PD) is the project’s scheduled duration and Actual Duration
(AD) is its actual duration at completion. To assess the
project’s cost performance (efficient use of BAC), Cost
Performance Index (CPI=EV/AC) is used. To measure the
project’s schedule progress, Schedule Performance Index
(SPI=EV/PV) is applied.
Finally, Cost Estimate at Completion (EAC($)) is the forecast
that represents the final cost of a project. Our study uses ML
to obtain a more accurate EAC($), and the index-based
formula in (1) is used as a benchmark. We compare our
EAC($) results with the ones calculated by (1).
($)=+− (1)
The linear model (1) is selected as the benchmark following
the study of Batselier and Vanhoucke (2015b), who conducted
a comparative analysis of eight index-based EAC($) models.
They used the EVM data of 51 real projects and associated
simulations for comparison. Based on the accuracy results,
measured by Mean Absolute Percentage Error (MAPE)
(defined next in the paper), they found that the model by (1)
showed dominance over the other seven models and produced
the most accurate EAC($) estimates.
2.2 Brief review of ML applications for project cost
forecasting
First, we note that ML models have not been extensively
applied in project monitoring and control. Only a few studies
developed ML models, specifically for project cost forecasting
during the project execution phase. Table 1 provides a
summary of these studies with a brief description of their
models and contribution to the EVM body of knowledge.
Among the first implementations, Pewdum et al. (2009)
developed an Artificial Neural Network (ANN) model based
on Backpropagation to improve the accuracy of duration and
cost estimates. They integrated numerous variables. Among all
the variables, the traffic volume, weather conditions, contract
duration, construction budget, percent complete of the planned
work, and percent complete of the actual work performed were
the most influential. Their model produced accurate EAC($)
when applied to highway construction projects in Thailand.
Table 1. A summary of the reviewed studies on ML
applications for project forecasting
Study
Description
Contribution to
EVM
Pewdum et
al. (2009)
Several project perfor-
mance factors are inte-
grated into the ANN-
based Backpropagation
model
Duration and
cost forecasting
during project
execution
Narbaev and
De Marco
(2014a)
Supervised regression
approach that integrates
the EVM cost data
through the Gompertz
Growth modeling
Cost forecasting
during project
execution
Elmousalami
(2021)
Numerous ML meth-
ods, including the En-
semble-based, are com-
pared using Fuzzy
Logic
Cost estimation
during project
planning
Ottaviani
and De
Marco
(2022)
Multiple linear regres-
sion model is proposed
using the EVM cost
data as independent (in-
put) variables
Cost forecasting
during project
execution
Natarajan
(2022)
The ANN and Refer-
ence class forecasting
approaches are inte-
grated to produce prob-
abilistic estimates
Duration and
cost forecasting
during project
planning and
execution
Wauters and
Vanhoucke
(2016)
The four ML tech-
niques are compared to
produce more accurate
duration estimates
Duration fore-
casting during
project execu-
tion
The current
study
The ANN-based Long-
Short Term Memory
model that uses the
EVM-based CPI and
SPI metrics and their
derivatives
Cost forecasting
during project
execution

3288 Tolgaİnanetal./IFACPapersOnLine55-10(2022)3286–3291
Narbaev and De Marco (2014a) adopted a Supervised
nonlinear regression approach based on the Gompertz Growth
model. Applying their model to nine construction projects, the
authors compared the accuracy of their estimates with the ones
produced by implementing the CPI-integrated index-based
model. As their model fits better to the S-shaped curve
observed in project cost growth, they obtained more accurate
cost estimates.
Elmousalami (2021) integrated ML algorithms into the cost
estimation efforts at the project development stage. Using the
Fuzzy Logic, they embedded the uncertainty factors in their
ML models and showed that the Ensemble methods were
superior in predicting performance.
Recently, Ottaviani and De Marco (2022) developed a multiple
linear regression model to assess the impact of input variables
(CPI, original cost forecast, and percent of work performed)
and improve the model fitting to the project’s real CAC value.
Using the data of 29 real-life projects, they showed that their
model with the three variables provided higher accuracy and
lower variance in EAC($) estimates.
Natarajan (2022) proposed a comprehensive model that
integrated Reference class forecasting (the outside view of a
project) and ML (the inside view from the project data) to
improve schedule and budget planning and control. Using the
cost data of 106 and the schedule data of 130 oil and gas
projects, the author showed a higher predictive capability of
the ML approach in predicting the most likely cost and
schedule overruns in projects.
To forecast the project duration, Wauters and Vanhoucke
(2016) applied Decision Tree, Bagging, Random Forest, and
Boosting techniques. They compared their forecasting results
with the ones by the conventional models (based on linear
performance indexes). Using artificial project data, they
showed that ML approaches had more accurate predicting
capabilities than the traditional index-based methods.
Finally, we refer to some review studies. Willems and
Vanhoucke (2015) examined the EVM methods and some ML
applications in project control. Hashemi et al. (2020) discussed
the ML applications for project cost forecasting. Ulusoy and
Hazir (2021) listed many interesting application areas of ML
in PM.
3. METHODOLOGY
3.1 Model
ML has been increasingly used in many fields, from computer
vision to biometric recognition, from advertising to the defense
industry. In the literature, ML approaches are classified into
Supervised learning, Unsupervised learning, Semi-supervised
learning, Reinforcement learning, and Dimensionality
reduction (e.g., Panda et al., 2021).
Unsupervised learning methods use input data, mainly to find
out the regularity in data. On the other hand, Supervised
techniques use both input and output data. Depending on the
problem, output data can be real numbers, integers, or
categories. In our study, the cost figures (real numbers)
constitute the output data.
Therefore, in this study, we focus on Supervised ML as there
is output data. In this approach, the training phase is crucial as
the patterns between the input-output data are found. On the
other hand, the testing phase of the Supervised ML generates
outputs following the input-output patterns determined during
the training phase. In Supervised ML, approaches can be
grouped into classification and regression subcategories. The
classification algorithms generate discrete or categorical
outputs. The regression algorithms generate continuous
outcomes. Time-sequence regression is the type of Supervised
ML that we implement in this study.
To explain how we used the time-sequence regression in our
study, we describe our approach including the processes used
in the ML training and testing phases. Fig. 1 presents how
these two phases are used within our prediction algorithm. We
learn the suitable ML model in the training phase, and the
learned ML model is used as a predictor in the testing phase.
We employ the ML model of a recurrent ANN type, namely
Long-Short Term Memory (LSTM). The LSTM networks are
suitable for sequence-to-sequence regression problems. We
refer to Hochreiter and Schmidhuber (1997) and Greff et al.
(2016) for more information on the LSTM networks.
ML models require features (inputs) to make the prediction.
Therefore, we must define the features of our ML model. We
design a seven-dimensional feature vector. Six dimensions of
the feature vector consist of CPI and SPI metrics and their
moving average filtered versions (having window sizes of two
and three tracking points for each metric). The seventh and last
dimension of the input vector is the normalized time. The
normalized time is found by dividing AD by PD for a
particular tracking point. The output (predicted value) of the
ML model is the cost at completion.
The training-testing protocol we use for the ML is as follows.
We use 12 projects in the training phase and three projects in
the testing phase of our experiments. Projects in the training
and testing phase are randomly selected. We repeat the
experiment a hundred times for each of the three projects,
covering the training and testing phases. Therefore, all projects
are used in both training and testing phases. Accordingly, the
results are reported independently of the training and testing
sets.
Figure 1. The training
and testing phases of the proposed ML
approach.

 Tolgaİnanetal./IFACPapersOnLine55-10(2022)3286–3291 3289
Narbaev and De Marco (2014a) adopted a Supervised
nonlinear regression approach based on the Gompertz Growth
model. Applying their model to nine construction projects, the
authors compared the accuracy of their estimates with the ones
produced by implementing the CPI-integrated index-based
model. As their model fits better to the S-shaped curve
observed in project cost growth, they obtained more accurate
cost estimates.
Elmousalami (2021) integrated ML algorithms into the cost
estimation efforts at the project development stage. Using the
Fuzzy Logic, they embedded the uncertainty factors in their
ML models and showed that the Ensemble methods were
superior in predicting performance.
Recently, Ottaviani and De Marco (2022) developed a multiple
linear regression model to assess the impact of input variables
(CPI, original cost forecast, and percent of work performed)
and improve the model fitting to the project’s real CAC value.
Using the data of 29 real-life projects, they showed that their
model with the three variables provided higher accuracy and
lower variance in EAC($) estimates.
Natarajan (2022) proposed a comprehensive model that
integrated Reference class forecasting (the outside view of a
project) and ML (the inside view from the project data) to
improve schedule and budget planning and control. Using the
cost data of 106 and the schedule data of 130 oil and gas
projects, the author showed a higher predictive capability of
the ML approach in predicting the most likely cost and
schedule overruns in projects.
To forecast the project duration, Wauters and Vanhoucke
(2016) applied Decision Tree, Bagging, Random Forest, and
Boosting techniques. They compared their forecasting results
with the ones by the conventional models (based on linear
performance indexes). Using artificial project data, they
showed that ML approaches had more accurate predicting
capabilities than the traditional index-based methods.
Finally, we refer to some review studies. Willems and
Vanhoucke (2015) examined the EVM methods and some ML
applications in project control. Hashemi et al. (2020) discussed
the ML applications for project cost forecasting. Ulusoy and
Hazir (2021) listed many interesting application areas of ML
in PM.
3. METHODOLOGY
3.1 Model
ML has been increasingly used in many fields, from computer
vision to biometric recognition, from advertising to the defense
industry. In the literature, ML approaches are classified into
Supervised learning, Unsupervised learning, Semi-supervised
learning, Reinforcement learning, and Dimensionality
reduction (e.g., Panda et al., 2021).
Unsupervised learning methods use input data, mainly to find
out the regularity in data. On the other hand, Supervised
techniques use both input and output data. Depending on the
problem, output data can be real numbers, integers, or
categories. In our study, the cost figures (real numbers)
constitute the output data.
Therefore, in this study, we focus on Supervised ML as there
is output data. In this approach, the training phase is crucial as
the patterns between the input-output data are found. On the
other hand, the testing phase of the Supervised ML generates
outputs following the input-output patterns determined during
the training phase. In Supervised ML, approaches can be
grouped into classification and regression subcategories. The
classification algorithms generate discrete or categorical
outputs. The regression algorithms generate continuous
outcomes. Time-sequence regression is the type of Supervised
ML that we implement in this study.
To explain how we used the time-sequence regression in our
study, we describe our approach including the processes used
in the ML training and testing phases. Fig. 1 presents how
these two phases are used within our prediction algorithm. We
learn the suitable ML model in the training phase, and the
learned ML model is used as a predictor in the testing phase.
We employ the ML model of a recurrent ANN type, namely
Long-Short Term Memory (LSTM). The LSTM networks are
suitable for sequence-to-sequence regression problems. We
refer to Hochreiter and Schmidhuber (1997) and Greff et al.
(2016) for more information on the LSTM networks.
ML models require features (inputs) to make the prediction.
Therefore, we must define the features of our ML model. We
design a seven-dimensional feature vector. Six dimensions of
the feature vector consist of CPI and SPI metrics and their
moving average filtered versions (having window sizes of two
and three tracking points for each metric). The seventh and last
dimension of the input vector is the normalized time. The
normalized time is found by dividing AD by PD for a
particular tracking point. The output (predicted value) of the
ML model is the cost at completion.
The training-testing protocol we use for the ML is as follows.
We use 12 projects in the training phase and three projects in
the testing phase of our experiments. Projects in the training
and testing phase are randomly selected. We repeat the
experiment a hundred times for each of the three projects,
covering the training and testing phases. Therefore, all projects
are used in both training and testing phases. Accordingly, the
results are reported independently of the training and testing
sets.
Figure 1. The training and testing phases of the proposed ML
approach.
The evaluation criteria to assess the accuracy of our model’s
cost estimate for a given project is the percentage error (the
percent difference between the cost estimate and the actual
cost of a project). We find the absolute average of these errors
for all the projects in our dataset and measure this average with
the Mean Absolute Percentage Error (MAPE), as in
 
󰈅󰇛󰇜
 󰈅󰇛󰇜


Where t=1,…, n is the number of tracking periods for a project.
3.2 Dataset
We use the actual project data shared by the Operations
Research & Scheduling Research Group of Ghent University
(ORSRG, 2022; Batselier and Vanhoucke, 2015a). This
database includes EVM data of 133 projects that have been
executed and completed in different industries. The dataset
mainly constitutes construction projects. Considering the
database structure, we limit our scope only to construction
projects, and our final dataset included the EVM data of 41
real-life completed projects.
The total duration and cost data of the 41 construction projects
extracted from the database are shown in Fig. 2. The projects
have a large range of durations and budgets. Considering this
variance, we chose the projects within a specific range. For the
budget, we kept the upper limit to 3 million Euros. For the
duration, we chose the projects with a maximum duration of
150 days and a minimum of four tracking points. The projects
that fall in these ranges have some similarities, but the others
are very small or big projects and quite different in project
characteristics and resources. By setting budget and time
limits, we generated a project pool of 15 projects. We
randomly selected 12 projects for the training set, and the
remaining three projects were reserved for testing.
4. SUMMARY of the RESULTS
Our results show that in 75.33% of the projects tested, the
MAPE (2) obtained using our ML model was smaller than that
obtained with the traditional index-based model (1). We found
the difference between MAPEs and provide its results as a
histogram in Fig. 3.
A positive difference in MAPE in this histogram shows a
smaller MAPE of our ML method. The negative difference
indicates the projects where our ML model produced a larger
MAPE than the conventional index-based model. About
50.00% of 75.33% projects tested have a MAPE difference of
about 1.00%. Even though this is a negligible difference in
EAC($) estimate’s accuracy between the two models, we note
that the proposed model has a feature to learn from the given
EVM data. This is because the EAC($) estimates calculated in
the testing phase followed the input-output patterns of the
EVM data of the projects analyzed in the training phase.
Following this, during the training phase, the cost-related
EVM data was utilized to build the proposed ML algorithm
using LSTM network. Our ML model evaluated this input data
repeatedly until learning its cost growth pattern (behavior).
5. CONCLUSION
Cost overrun is a common problem in projects undertaken in
various industries. To deal with this common problem, project
managers opt for continuously monitoring the use of the
project budgets. Many of them try to produce accurate cost
estimates using the traditional EVM methods. These methods
are mainly based on cost and schedule performance indexes
which are linear. However, projects' budget acquisition and
cost growth patterns are nonlinear and resemble an S-shape.
Therefore, such methods have the inherent limitations in
providing more reliable and accurate cost estimates that reflect
the real cost growth behavior. Also, whatever the approach
followed or the method implemented, having accurate cost
estimates is critical to completing the projects successfully and
maintaining loyal relationships with project stakeholders.
Considering this limitation of the existing index-based models
and the importance of having accurate cost estimates, in this
study, we developed an ML algorithm for estimating the total
project cost more accurately. We employed a Supervised ML
model based on the LSTM protocol to forecast EAC($). The
EVM data of 41 real completed projects validated the proposed
approach. The training phase of our approach with 12 projects
allowed us to learn from the given dataset the patterns that
characterized the changes in the project cost. We used the
seven-dimensional feature vector that considered EVM
metrics like CPI and SPI and their moving averages and the
Figure 2. The cost and duration plot for 41 projects.
Figure 3. The MAPE difference between the proposed ML model
and the EVM index-based model.

3290 Tolgaİnanetal./IFACPapersOnLine55-10(2022)3286–3291
normalized time as a predictor. Based on this, we used the
learned patterns to calculate EAC($). In the testing phase, we
validated our approach on three projects with an associated
hundred experiments for each project. We compared our
approach’s EAC($) accuracy results with the ones computed
using the widely used index-based model in practice (1).
Overall, our model produced more accurate EAC($) results in
75.33% of project cases.
We acknowledge the following limitations that can potentially
be addressed in future research. First, we conducted the
experiments using a small dataset. We intend to extend the
current research using a larger pool of projects and evaluate
the model using additional forecasting criteria such as stability
and timeliness of EAC($), in addition to the accuracy. Second,
we will also work with projects from different industries, not
only construction. However, the initial results obtained with
the proposed ML approach are promising. The proposed
method can be combined with other forecasting techniques to
improve the solutions further.
ACKNOWLEDGMENTS
This research was funded by the Science Committee of the
Ministry of Education and Science of the Republic of
Kazakhstan (Grant No. AP09259049).
REFERENCES
Anbari, F.T. (2003). Earned value project management
method and extensions. Project Management Journal,
34(4), 12–23.
Barraza, G.A., Back, W.E., and Mata, F. (2004). Probabilistic
forecasting of project performance using stochastic S
curves. Journal of Construction Engineering and
Management, 130(1).
Batselier, J. and Vanhoucke, M. (2015a). Construction and
evaluation framework for a real-life project database,
International Journal of Project Management, 33(3),
697–710.
Batselier, J. and Vanhoucke, M. (2015b). Empirical evaluation
of earned value management forecasting accuracy for
time and cost. Journal of Construction Engineering and
Management, 141(11), 05015010.
Cacavid, J.P.U., Lamouri, S., Grabot, B., and Fortin, A.
(2019). Machine Learning in Production Planning and
Control: A Review of Empirical Literature. IFAC
PapersOnLine, 52(13), 385–390.
Chen, Z., Demeulemeester, E., Bai, S., and Guo, S. (2020). A
Bayesian approach to set tolerance limits for a statistical
project management. International Journal of
Production Research, 58(10), 3150-3163.
De Marco, A., Rosso, M., and Narbaev, T. (2016). Nonlinear
cost estimates at completion adjusted with risk
contingency. Journal of Modern Project Management,
4(2), 24–33.
Elmousalami, H.H. (2020). Comparison of Artificial
Intelligence techniques for project conceptual cost
prediction: A case study and comparative analysis. IEEE
Transactions on Engineering Management, 68(1), 183-
196.
Greff, K., Srivastava, R.K., Koutník, J., Steunebrink, B.R., and
Schmidhuber, J. (2016). LSTM: A search space odyssey.
IEEE Transactions on Neural Networks and Learning
Systems, 28(10), 2222-2232.
Hashemi, S.T., Ebadati, O.M., and Kaur, H. (2020). Cost
estimation and prediction in construction projects: a
systematic review on machine learning techniques. SN
Applied Sciences, 2, 1703.
Hochreiter, S. and Schmidhuber, J. (1997). Long short-term
memory. Neural Computation, 9(8), 1735-1780.
Humphreys, G.C. (2018). Project management using Earned
value. 4th edition. Humphreys & Associates, Inc.
IPMA. (2020). Report on Artificial Intelligence impact in
Project Management. International Project Management
Association. Amsterdam, The Netherlands.
Kim, B.C. and Reinschmidt, K.F. (2011). Combination of
project cost forecasts in earned value management.
Journal of Construction Engineering and Management,
137(11), 958–966.
Lipke, W., Zwikael, O., Henderson, K., and Anbari, F. (2009).
Prediction of project outcome: the application of
statistical methods to earned value management and
earned schedule performance indexes. International
Journal of Project Management, 27(4), 400–407.
Mahmoudi, A., Bagherpour M., and Javed, S.A. (2021). Grey
earned value management: Theory and applications.
IEEE Transactions on Engineering Management, 68(6),
1703-1721.
Munir, M. (2019). How Artificial Intelligence can help project
managers. Global Journal of Management And Business
Research, 19(4), 1-8.
Narbaev, T. and De Marco, A. (2014a). An Earned Schedule-
based regression model to improve cost estimate at
completion, International Journal of Project
Management, 32(6), 1007-1018.
Narbaev, T. and De Marco, A. (2014b). Combination of
growth model and earned schedule to forecast project
cost at completion. Journal of Construction Engineering
and Management, 140(1), 04013038.
Natarajan, A. (2022). Reference class forecasting and Machine
Learning for improved offshore oil and gas megaproject
planning: Methods and application. Project Management
Journal, 53(OnlineFirst), 1-29.
Ong, S. and Uddin, S. (2020). Data science and Artificial
Intelligence in project management: The past, present

 Tolgaİnanetal./IFACPapersOnLine55-10(2022)3286–3291 3291
normalized time as a predictor. Based on this, we used the
learned patterns to calculate EAC($). In the testing phase, we
validated our approach on three projects with an associated
hundred experiments for each project. We compared our
approach’s EAC($) accuracy results with the ones computed
using the widely used index-based model in practice (1).
Overall, our model produced more accurate EAC($) results in
75.33% of project cases.
We acknowledge the following limitations that can potentially
be addressed in future research. First, we conducted the
experiments using a small dataset. We intend to extend the
current research using a larger pool of projects and evaluate
the model using additional forecasting criteria such as stability
and timeliness of EAC($), in addition to the accuracy. Second,
we will also work with projects from different industries, not
only construction. However, the initial results obtained with
the proposed ML approach are promising. The proposed
method can be combined with other forecasting techniques to
improve the solutions further.
ACKNOWLEDGMENTS
This research was funded by the Science Committee of the
Ministry of Education and Science of the Republic of
Kazakhstan (Grant No. AP09259049).
REFERENCES
Anbari, F.T. (2003). Earned value project management
method and extensions. Project Management Journal,
34(4), 12–23.
Barraza, G.A., Back, W.E., and Mata, F. (2004). Probabilistic
forecasting of project performance using stochastic S
curves. Journal of Construction Engineering and
Management, 130(1).
Batselier, J. and Vanhoucke, M. (2015a). Construction and
evaluation framework for a real-life project database,
International Journal of Project Management, 33(3),
697–710.
Batselier, J. and Vanhoucke, M. (2015b). Empirical evaluation
of earned value management forecasting accuracy for
time and cost. Journal of Construction Engineering and
Management, 141(11), 05015010.
Cacavid, J.P.U., Lamouri, S., Grabot, B., and Fortin, A.
(2019). Machine Learning in Production Planning and
Control: A Review of Empirical Literature. IFAC
PapersOnLine, 52(13), 385–390.
Chen, Z., Demeulemeester, E., Bai, S., and Guo, S. (2020). A
Bayesian approach to set tolerance limits for a statistical
project management. International Journal of
Production Research, 58(10), 3150-3163.
De Marco, A., Rosso, M., and Narbaev, T. (2016). Nonlinear
cost estimates at completion adjusted with risk
contingency. Journal of Modern Project Management,
4(2), 24–33.
Elmousalami, H.H. (2020). Comparison of Artificial
Intelligence techniques for project conceptual cost
prediction: A case study and comparative analysis. IEEE
Transactions on Engineering Management, 68(1), 183-
196.
Greff, K., Srivastava, R.K., Koutník, J., Steunebrink, B.R., and
Schmidhuber, J. (2016). LSTM: A search space odyssey.
IEEE Transactions on Neural Networks and Learning
Systems, 28(10), 2222-2232.
Hashemi, S.T., Ebadati, O.M., and Kaur, H. (2020). Cost
estimation and prediction in construction projects: a
systematic review on machine learning techniques. SN
Applied Sciences, 2, 1703.
Hochreiter, S. and Schmidhuber, J. (1997). Long short-term
memory. Neural Computation, 9(8), 1735-1780.
Humphreys, G.C. (2018). Project management using Earned
value. 4th edition. Humphreys & Associates, Inc.
IPMA. (2020). Report on Artificial Intelligence impact in
Project Management. International Project Management
Association. Amsterdam, The Netherlands.
Kim, B.C. and Reinschmidt, K.F. (2011). Combination of
project cost forecasts in earned value management.
Journal of Construction Engineering and Management,
137(11), 958–966.
Lipke, W., Zwikael, O., Henderson, K., and Anbari, F. (2009).
Prediction of project outcome: the application of
statistical methods to earned value management and
earned schedule performance indexes. International
Journal of Project Management, 27(4), 400–407.
Mahmoudi, A., Bagherpour M., and Javed, S.A. (2021). Grey
earned value management: Theory and applications.
IEEE Transactions on Engineering Management, 68(6),
1703-1721.
Munir, M. (2019). How Artificial Intelligence can help project
managers. Global Journal of Management And Business
Research, 19(4), 1-8.
Narbaev, T. and De Marco, A. (2014a). An Earned Schedule-
based regression model to improve cost estimate at
completion, International Journal of Project
Management, 32(6), 1007-1018.
Narbaev, T. and De Marco, A. (2014b). Combination of
growth model and earned schedule to forecast project
cost at completion. Journal of Construction Engineering
and Management, 140(1), 04013038.
Natarajan, A. (2022). Reference class forecasting and Machine
Learning for improved offshore oil and gas megaproject
planning: Methods and application. Project Management
Journal, 53(OnlineFirst), 1-29.
Ong, S. and Uddin, S. (2020). Data science and Artificial
Intelligence in project management: The past, present
and future. Journal of Modern Project Management,
7(4), 123–456.
ORSRG (2022). Real data. Operations Research & Scheduling
Research Group. Ghent University. Available at
https://www.projectmanagement.ugent.be/research/data
/realdata
Ottaviani, F.M. and De Marco, A. (2022). Multiple linear
regression model for improved project cost forecasting.
Procedia Computer Science, 196(2022) 808–815.
Panda, S.K., Mishra, V., Balamurali, R., and Elngar, A.A.
(2021). Artificial Intelligence and Machine Learning in
business management: Concepts, challenges, and case
studies. 1st edition. CRC Press Taylor&Francis Group.
Florida, US.
Pellerin, R. and Perrier, N. (2019). A review of methods,
techniques and tools for project planning and control.
International Journal of Production Research, 57(7),
2160–2178.
Pewdum, W, Rujirayanyong, T., and Sooksatra, V. (2009).
Forecasting final budget and duration of highway
construction projects. Engineering, Construction, and
Architectural Management, 16(6), 544–557.
PMI. (2019). The standard for Earned Value Management. 2nd
edition. Project Management Institute (PMI). Newtown
Square, PA.
Rai, R., Tiwari, M.K., Ivanov, D., and Dolgui, A. (2021).
Machine Learning in manufacturing and industry 4.0
applications. International Journal of Production
Research, 59(16), 4773-4778.
Ulusoy, G. and Hazir, Ö. (2021). Recent developments and
some promising research areas. In Ulusoy, G. and Hazır,
Ö. An introduction to project modeling and planning,
457-469. Springer Texts in Business and Economics.
Springer, Cham.
Warburton, R.D.H. and Cioffi, D.F. (2016). Estimating a
project’s earned and final duration. International
Journal of Project Management. 34 (8), 1493–1504.
Wauters, M. and Vanhoucke, M. (2016). A comparative study
of Artificial Intelligence methods for project duration
forecasting. Expert Systems with Applications, 46, 249-
261.
Willems, L.L. and Vanhoucke, M. (2015). Classification of
articles and journals on project control and Earned Value
Management. International Journal of Project
Management, 33(7), 1610–1634.