A Novel Approach to Predict Chronic Kidney Disease using Machine Learning Algorithms

Full text

A Novel Approach to Predict Chronic Kidney Disease
using Machine Learning Algorithms
Bhavya Gudeti
School of Computer Science and
Engineering
Vellore Institute of Technology,
Chennai, 600127, Tamilnadu, India.
bhavyagudeti@gmail.com
Terrance Frederick Fernandez
[0000-0002-7317-3362]*
Rajiv Gandhi College of Engineering
and Technology, Puducherry, India
frederick@pec.edu
Shashvi Mishra
School of Computer Science and
Engineering
Vellore Institute of Technology,
Chennai, 600127,
Tamilnadu, India.
shashvimishra@gmail.com
Amit Kumar Tyagi [0000-0003-2657-8700]*
School of Computer Science and
Engineering,
Vellore Institute of Technology,
Chennai, 600127, Tamilnadu, India.
amitkrtyagi025@gmail.com
Shaveta Malik
Terna College of Engineering,
University of Mumbai ,
Maharashtra, India
shavetamalik687@gmail.com
Shabnam Kumari
Anumit Academy of Research and
Research Network, India
shabnam.kt25@gmail.com
Abstract—A staggering 63,538 cases have been registered,
according to India's health statistics upon Chronic Kidney
Disease (CKD). The average age of nephropathy for humans lies
between 48-70 years. CKD is more prevalent among males than
females. Bitterly, our Nation rank among top 17 countries in
CKD since 2015 which is characterized by a gradual loss of
excretory organ performance over time. Earlier detection of the
illness followed by treatment could keep this dreaded disease at
the shore. Machine Learning, is making sensible applications in
areas such as analyzing medical science outcomes, sleuthing
fraud etc. For the prediction of chronic diseases various machine
learning algorithms are implemented.
Our main aim is to differentiate the performance of various
machine learning algorithms primarily based on its accuracy. In
this work we idolized Rcode to compare their performances. The
pivotal purpose of this study is to analyze the Chronic Kidney
Disease dataset and conduct CKD and Non CKD classification
cases.
Keywords— MachineLearning, Chronic Kidney Disease,
Classification, Accuracy, LogisticRegression, Support Vector
Machine
I. INTRODUCTION
Way back in 1950s, the communication among the human
were predominantly oral. However as technology progressed
since, mankind were obsessed with the technology. The million
dollar query remains, "Why Humans are more obsessed with
technology?”. The response is straight-forward. Demand rise
in manufacturing accelerates surplus data in trade growth,
product, business perspective and sales. These days industries
like automation, aerospace, health care, etc., are operating in
communication of Machines or interconnection of Internet of
Things (IoTs). These IoTs devices (in interconnection) are
manufacturing heaps of knowledge that is required to be
analyzed with efficiency through efficient and fashionable
tools/approaches. Current available ancient tools don’t seem to
be enough to analyze huge volumes of data.
Clustering may be thought as an assortment of objects into
clusters that are similar in nature. The cluster/group contains
objects that mimic each other, while the objects in the other
ones are dissimilar. The application is widely applicable in
applications like Marketing, World Wide Web (WWW),
Earthquake Studies, Aerospace, Biology, Insurance, etc. On
another hand, if the information renders with categorizes/ class
labels, and then classification technique is employed to
categorize the given information into number of classes/
categories based on their similarities. The various applications
of classification are speech and handwriting recognition,
Identification of biometric, classification of documents, etc.
Association Rule Mining (ARM) is: if-then statements
facilitate to indicate the relationships between data items
among transactional databases. Further, Regression (or linear
regression) is employed to seek out the relationship between
two continuous variables. One variable is termed as predictor
or independent variable and other is dependent or response
variable. Outlier detection is outlined as “The method of
distinguishing the extreme points within the data. It is a branch
of data mining.” These all algorithms (discussed above) are a
part of Data mining/ Machine Learning/ Computer Vision.
In the human body, the kidney is instrumental in absorbing
and discharging all the toxic and unessential materials,
typically wastes, from the body through egesting and excretion
process. In India, there are approximately one million cases of
Chronic Kidney Disease (CKD) every year. It is dangerous to
kidney and it produces gradual loss in kidney functionality.
Nevertheless, it is unpredictable because its symptoms grow
gradually and are not unique to the disorder, it is important to
detect CKD at its early stage. Kidneys filter wastes and excess
fluids from the blood that are then excreted in excrement. In
the early stages of CKD, we will have a few signs or
symptoms.
In healthcare organization, Classification is one in all the
foremost usually used ways of machine learning. The

classification model shows the class of result for each data
point.
The classifying methods are Decision Tree, Support Vector
Machine, K-Nearest Neighbor, Naïve Bayes classifier, and
Neural Networks. KNN is used to visualize at the relationship
between different CKD risk factors, in order to predict the
disease at an early stage. Machine Learning is a growing phase
dealing with the study of a huge variable data and it is grown
from the study of pattern (speech and handwriting) recognition
and computational learning theory in Artificial Intelligence
having numerous methods, algorithms, and techniques to
analyze and predict the data. Machine Learning techniques
have proved huge success in detection and recognition of many
essential diseases in medical science’s point of view. Machine
learning would thus be useful for predicting whether the patient
has CKD or not in this question. By using old CKD patient
data to train predictive model, Machine Learning does so.
A. Analysis of Chronic Kidney Disease (CKD)
It makes its way as a ground-breaking and actual channel
that liberates the body from squander and parlous substances
and return supplements, amino acids, insulin, hormones and
different basic substances to the circulatory framework.
Incidentally things will flip out gravely, however. “Chronic
Kidney Disease (CKD) is used at some stage in the world to
suggest to any variety of nephritis that returns. “Infection”
incorporates any deviation from the urinary organ structure or
limit customary, paying very little heed to whether or not it is
most likely going to create a man feel unwell or manufacture
complexities. It is a typical issue that may influence anybody at
any age. It’s assessed that just about 3 million people within
the United Kingdom are at risk of CKD. A combination of
totally different conditions that usually place as train on the
kidneys works on CKD.
Hence, the manuscript is organized as follows: Section II
mentions about related work upon this research topic. Section
III, discusses the Proposed System to classify Chronic Kidney
Disease (CKD). Section IV describes the information regarding
the dataset used and transient introduction regarding the
attributes. Section V deals with the machine learning
algorithms, code and its results for variable measures and
therefore the corresponding output obtained in each
classification algorithm. Further, section VI discusses about an
open discussion about current view, results about chronic
disease. Section VII finally discusses the conclusion of the
research work alongside with the attribute improvement.
II. `RELATED WORK
There are diverse researchers who have worked with the
assistance of several different classification algorithms on CKD
prediction. All those had their model performance expected.
Gunarathne, W.H.S.D. [1] compared the effects of divergent
models. Finally, they concluded that the Multiclass Decision
forest algorithm provides plentiful precision for the 14-attribute
(reduced) data set. S.Dilli Arasu and Dr. R. Thirumalaiselvi [2]
worked on missing values in a Chronic Kidney Disease dataset.
They deduced that the missing values in the dataset can not
only reduce the model's accuracy but also the effects of the
prediction. By patterning a numerical method on stages of
Chronic Kidney Disease, they found a solution to this issue and
by doing so; they stood up with unknown values. They
substituted the missing values with those recalculated ones.
In discovering Chronic Kidney Disease using machine
learning algorithms, Asif Salekin and John stankovic[3] used
novel approach. They got findings on a dataset consisting of
400 records and 25 attributes resulting in a patient prone to
CKD or not. In order to achieve results, they used KNN,
random forest and Neural Network algorithms. They used
wrapper methodology for feature reduction which finds CKD
with high accuracy.
12 different classification algorithms on various datasets
were tested by Sahil Sharma, Vinod Sharma, and Atul Sharma
[4], each with 400 records and 24 attributes. They compared
their expected outcomes with actual results in order to
determine predictive accuracy. They used metrics such as
precision, sensitivity, accuracy and specificity for measuring
the performance of the classifiers. Note that Chronic Kidney
Disease (CKD) is not uncommon.
However, a lot of correct information regarding risk for
progression to nephropathy is direly needed for clinical
selections concerning testing, treatment and referral. Hence,
this section highlighted upon the state of art in the field of
CKD. Interestingly, the further section would discuss our work
in detail.
III. PROPOSED SYSTEM
The proposed system deals with the detection of Chronic
Kidney disease. The healthcare systems generate colossal data.
Thus, it is obligatory to use this data productively to analyze,
predict, and to treat an explicit disease. A classification model
offers some solution from determined values. In classification
type, we have a tendency to expect fewer or lots of input to
predict values of their outcomes. In a supervised machine
learning algorithms, the classification algorithm uses the
training dataset. Classification predicts the categorical class
labels in the data.
The research work tries to present a machine learning
framework for information discovery on the Chronic Kidney
Disease dataset. To classify the disease at puerility, three
machine learning algorithms are used, namely Logistic
Regression, Support Vector Machine, and K-Nearest
Neighbors. The molarity of every algorithm is inspected. Our
proposed model combines the Support Vector Machine,
Logistic Regression and K-Nearest Neighbours (KNN) as
mentioned in Fig.1.
This section snapshot our system that has been proposed in
this paper. Now, next section will discuss the datasets that are
being employed and introduce the table that shows the
attributes and description on the same.

Fig.1. Proposed Model using Various Classification Algorithms
IV. `DATASET USED
The proposed framework uses the UCI Machine Learning
Repository dataset called Chronic Kidney Disease (CKD) that
has 25 attributes, out of which, 11 are numerical and 14 are
nominal. Entire 400 instances of the dataset are used for
training to predict machine learning algorithms. In 400
instances, 250 are labeled as Chronic Kidney Disease (CKD)
and 150 are labeled as Non Chronic Kidney disease. The
attributes present in the data set are bacteria, sodium, age,
Hemoglobin, Diabetes Mellitus, Classification, Appetite,
Coronary Artery Disease, Blood Pressure, Pus cell, Anemia,
Pedal Edema, Sugar, White Blood Cell Count, Hypertension,
Red Blood Cell Count, Potassium, Specific Gravity, Pus cell
chumps, Packed Cell Volume, Albumin, Serum Creatinine,
Red Blood Cells, Blood Urea, and Blood Glucose Random.
The dataset that is taken is divided into two groups, one for
testing the samples and another for training the samples. The
ratio for testing and training data is 30% and 70% respectively.
The data set used has been listed in table 1. The readers can
refer following URL [16] for collecting data. Now, next section
will discuss regarding the machine learning algorithms used to
classify CKD.
TABLE 1 :DATA SET USED
S.No
Attribute
Description about the attribute
1.
Bacteria(nominal)
ba – (present / not present)
2.
Sodium(numerical)
sod in mEq/L
3.
Age (numerical)
Person’sAgeinYears
4.
Haemoglobin (numerical)
Hemo in grams
5.
Diabetes Mellitus (nominal)
dm – ( yes / no)
6.
Class (nominal )
class – (ckd / notckd)
7.
Appetite (nominal)
appet – (good / poor)
8.
Coronary Artery Disease
(nominal)
CAD – (yes / no)
9.
Blood Pressure (numerical)
BP in mm/Hg
10.
Pus cell (nominal)
PC – (normal / abnormal)
11.
Anemia (nominal)
ane – (yes / no)
12.
Pedal Edema (nominal)
pe – (yes / no)
13.
Sugar (nominal)
su – (0/1/2/3/4/5)
14.
White Blood CellCount
(numerical )
Wc in cells/cumm
15.
Hypertension (nominal)
htn – (yes/no)
16.
Red Blood Cell Count
(numerical)
Rc in cells/cumm
17.
Potassium (numerical)
Pot in mEq/L
18.
Specific Gravity (nominal)
Sg -
(1.005/1.010/1.015/1.020/1.025)
19.
Pus Cell clumps (nominal)
pcc – (present / notpresent)
20.
Packed Cell Volume
(numerical)
P cv
21.
Albumin (nominal)
al – (0/1/2/3/4/5)
22.
Serum
Creatinine(numerical)
Sc in mgs/dl
23.
Red Blood Cells (nominal)
RBC – (normal/ abnormal)
24.
Blood Urea (numerical)
Bu in mgs/dl
25.
Blood Glucose Random
(numerical)
BGR in mgs/dl
V. SIMULATION RESULTS
This section describes about the simulation results that
are being used in the paper here.
A. Logistic Regression
Logistic Regression may be a calculation for order. As
per heaps of autonomous factors, the logic is 1/0, Yes/No,
True/False. It can be employed to access a paired answer. We
have a tendency to utilize the likelihood log as an
impoverished variable. Logistic Regression is used for the
classification problems in Machine Learning Algorithms. It is
a prophetic analysis algorithm and it is based mostly on the
concept of probability. It means that, given a certain factor,
logistic regression is used to predict an outcome that has two
values. The source code is exemplified in Table I and the
output in Fig.2. From them, we can deduce that the accuracy
of Logistic Regression is 0.7725
TABLE II : RCODE FOR LOGISTIC REGRESSION

ckd<- read.csv("C:/Users/bhavya/Desktop/ckd.csv")
ckd
ckd$Type<- NULL
head(ckd)
dim(ckd)
summary(ckd)
names(ckd)
contrasts(ckd$classification)
#Logistic Regression
glm.fit=glm(classification~age+bp+pcv+bu,
data=ckd,family=binomial)
summary(glm.fit)
#predict provides a vector of fitted probabilities.
glm.probab=predict(glm.fit,type="response")
glm.probab[1:20]
glm.predc=rep("ckd",400)
glm.predc[glm.probab>.5]="notckd"
table(glm.predc,ckd$classification)
mean(glm.pred==ckd$classification)
Fig.2. Output for Logistic Regression
B. Support Vector Machines (SVM)
For each relapse and grouping undertakings, Support
Vector Machine, curtailed as SVM, will be used. Multitude of
researchers favors it deeply as it provides unbelievable
accuracy with less power of activity. In ML, SVM support
vector systems are supervised models compatible with
learning. Support Vector Machine (SVM) offers a dual
platform for regression and classification. This can be used to
solve both linear problems and non-linear ones. This algorithm
uses a hyper plane to categorize the data points. Within this
SVM algorithm, each data point will be plotted as a point in n-
dimensional space, with a value of each attribute being the
value of a given coordinate. Classification can be
accomplished by searching for the right hyper-plane which
basically distinguishes between the two CKD and not CKD
groups. Table III presents the code behind SVM and from the
results in Fig.3, we can witness that the accuracy of SVM =
0.9925187
TABLE III : RCODE FOR SVM
#Generate a random number that is 70% of the total number of
rows in dataset.
ckd1 <- sample(1:nrow(ckd),0.7*nrow(ckd))
ckd.train<- ckd[ckd1,]
ckd.test<- ckd[-ckd1,]
set.seed(1)
ckd<-ckd[1:200,]
x=cbind.data.frame(ckd.train[,9:13])
y=ckd.train$classification
dataset=data.frame(x=x, y=as.factor(y))
library(e1071)## Support Vector Machine
svmfit=svm(y~., data=dataset, kernel="radial",gamma=1,
cost=1)
summary(svmfit)
svm.probs=predict(svmfit,type="response")
svm.probs[1:400]
svm.pred=rep("ckd",400)
svm.pred[svm.probs="notckd"]="notckd"
mean(svm.pred==ckd$classification)
Fig.3. Output for SVM
C. K-Nearest Neighbors Classification:
The sole performance of the K nearest neighbor classifier
algorithm is to predict the target variable by capturing the
nearest neighbor class. The nearest class will be known as the
target variable using the distance measures like Euclidean
distance.
Algorithm:
1. Initialize the parameter K.
2. Calculate the distance between the test sample and all
the training samples
3. Sort the distance in the ascending order.
4. Take K-nearest neighbors.
5. Gather the class of the nearest neighbor.
6. Here as we can see the accuracy in KNN = 0.7875
From the algorithm mentioned above, it is evident
that the results are better in Support Vector Machine. We
provide result with an accuracy of 0.9925187. Now, the
subsequent section will provide a conclusion regarding this
work in brief adding some future enhancement possibilities
with this work.
TABLE IV : SIMULATION RESULTS
Name of Classifier
Accuracy
Logistic Regression
0.7725
Support Vector Machine (SVM)
0.9925187
K-Nearest Neighbour
0.7875
`VI. AN OPEN DISCUSSION

Each classifier's results were evaluated using different
evaluation parameters, and cross-checked against over-fitting
with 10-fold cross-validation. The technique of nested cross-
validation has also helped to fine-tune the model parameters.
The tests will be carried out using the Python 3.3
programming language through the Jupyter Notebook web
application. Several Sciket-learning libraries were used, which
is a free machine learning system platform for Python.
Accuracy using F1-measurement, sensitivity, specificity and
Area under Curve (AUC) are the assessment measures
considered in this analysis. Each model produces different
outputs; depending on its parameter values.Thus with the GB
model we achieve the best efficiency in detection. This result
is better than the results obtained by using a multilayer
perceptron algorithm (MLP) single point split, seven
attributes, and a 98.4 percent F1 measurement. By contrast, a
98.0 per cent F1-measure was obtained with better efficiency
relative to study using RF and five apps.
Some limitations on the dataset used are, however,
important to this analysis. Second, the sample size (400
instances) is expected to be low and may affect the reliability
of the studies. Second, problem identification is another
dataset which has the same features to assess the performance
of the data sets. Also, the readers are suggested to read [17,
18, 19, 20, 21 and 22] to know more about artificial
intelligence, machine learning and deep learning techniques,
i.e., their scope in near future.
VII. CONCLUSION AND FUTURE WORK
Aimed to diagnose Chronic Kidney Disease (CKD) at an
earlier stage, this manuscript introduced a variety of machine
learning algorithms. The models obtained from CKD patients
are trained and authenticated with the mentioned input
parameters. Support Vector Machine, Logistic Regression and
knn are analyzed to conduct the study of CKD. The
performances of those algorithms were determined primarily
on the basis of precision. Our results exemplified that the
Support Vector Machine algorithm predicts Chronic Kidney
Disease better than Logistic Regression and K-Nearest
Neighbors within the narrow limits of this medical scenario.
The benefit of this approach is that the prediction process
takes far less time and helps doctors to initiate treatment at the
earliest for patients with CKD and further to classify larger
population of patients within shorter span. Because the dataset
used in this paper is tiny with 400 examples, we prefer to work
with larger datasets in the future or compare the results of this
dataset with a different dataset with the same. In addition, to
help minimise the incidence of CKD, we try to predict if a
person with this syndrome chances chronic risk factors such as
hypertension, family history of kidney failure and diabetes
using the appropriate dataset.
AUTHOR’S CONTRIBUTIONS
Shashvi and Bhavya have drafted this manuscript. Amit
Kumar Tyagi and Terrance Frederick Fernandez have
analyzed and approved this manuscript for publication.
CONFLICT OF INTEREST
The authors do not have any conflict concerning
publication of this manuscript.
REFERENCES
[1] L. Rubini, “Early stage of chronic kidney disease UCI machine learning
repository,”2015. [Online].
Available:http://archive.ics.uci.edu/ml/datasets/Chronic Kidney Disease.
[2] Asif Salekin, John Stankovic, "Detection of Chronic Kidney Disease and
Selectiing Important Predictive Attributes," Proc. IEEE International
Conference on Healthcare Informatics (ICHI), IEEE, Oct. 2016,
doi:10.1109/ICHI.2016.36.
[3] Q. Zhang and D. Rothenbacher, "Prevalence of chronic kidney disease
in population-based studies: systematic review," BMC Public Health,
vol. 8, (1), pp. 117, 2008.
[4] K. A. Padmanaban and G. Parthiban, "Applying Machine Learning
Techniques for Predicting the Risk of Chronic Kidney Disease," Indian
Journal of Science and Technology, vol. 9, (29), 2016.
[5] J. Xiao et al, "Comparison and development of machine learning tools in
the prediction of chronic kidney disease progression," Journal of
Translational Medicine, vol. 17, (1), pp. 119, 2019.
[6] Sahil Sharma, Vinod Sharma, Atul Sharma, “Performance Based
Evaluation of Various Machine Learning Classification Techniques for
Chronic Kidney Disease Diagnosis,” July18, 2016.
[7] GunarathneW.H.S.D, Perera K.D.M, Kahandawaarachchi K.A.D.C.P,
“Performance Evaluation on Machine Learning Classification
Techniques for Disease Classification and Forecasting through Data
Analytics for Chronic Kidney Disease (CKD)”, 2017 IEEE
17thInternational Conference on Bioinformatics and Bioengineering.
[8] S.Ramya, Dr.N.Radha, "Diagnosis of Chronic Kidney Disease Using
Machine Learning Algorithms," Proc. International Journal of
Innovative Research in Computer and Communication Engineering, Vol.
4, Issue 1, January 2016.
[9] S. A. Shinde and P. R. Rajeswari, “Intelligent health risk prediction
systems using machine learning: a review,” IJET, vol. 7, no. 3, pp.
1019– 1023, 2018.
[10] A.J. Aljaaf et al, "Early prediction of chronic renal
disorder mistreatment machine learning supported
by prognosticative analytics," in 2018 IEEE
Congress on organic process Computation (CEC), 2018.
[11] J.Xiao et al, "Comparison and development of machine learning tools in
the prediction of chronic renal disorder progression," Journal of
Translational drugs, vol. 17, (1), pp. 119, 2019.
[12] P. Yang et al, "A review of ensemble strategies in bioinformatics,
"Current Bioinformatics, vol. 5, (4), pp. 296-308, 2010.
[13] L.Deng et al, "Prediction of protein-protein interaction
sites mistreatment associate
ensemble methodology," BMC Bioinformatics, vol. 10, (1), pp. 426,
2009.
[14] M. Moslem and M. Pasha, "Survey of machine learning algorithms
fordisease diagnostic," Journal of Intelligent Learning Systems
andApplications, vol. 9, (01), pp. 1, 2017.
[15] S.Karamizadeh et al, "Advantage and disadvantage of support vector
machine practicality," in 2014 International Conference on laptop,
Communications, and management Technology (I4CT), 2014.
[16] http://archive.ics.uci.edu/ml/datasets/Chronic_Kidney_Disease
[17] Akshara Pramod, Harsh Sankar Naicker, Amit Kumar Tyagi, “Machine
Learning and Deep Learning: Open Issues and Future Research
Directions for Next Ten Years”, Book: Computational Analysis and
Understanding of Deep Learning for Medical Care: Principles, Methods,
and Applications, 2020, Wiley Scrivener, 2020.
[18] Tyagi, Amit Kumar and G, Rekha, Machine Learning with Big Data
(March 20, 2019). Proceedings of International Conference on
Sustainable Computing in Science, Technology and Management
(SUSCOM), Amity University Rajasthan, Jaipur - India, February 26-
28, 2019.
[19] Amit Kumar Tyagi, Poonam Chahal, “Artificial Intelligence and
Machine Learning Algorithms”, Book: Challenges and Applications for
Implementing Machine Learning in Computer Vision, IGI Global, 2020.

[20] Amit Kumar Tyagi, G. Rekha, “Challenges of Applying Deep Learning
in Real-World Applications”, Book: Challenges and Applications for
Implementing Machine Learning in Computer Vision, IGI Global 2020,
p. 92-118.
[21] Terrance Frederick Fernandez and M. Pradeep, "Multi-level Predictive
with Training Framework (MP with TF) for ranking machine learning
algorithms", IEEE proceeding of 4th International conference on I-
SMAC (IoT in Social, Mobile, Analytics and Cloud (I-SMAC 2020),
pp.697-703, ISBN: 978-1-7281-5464-0/20, 7th to 9th October 2020,
SCAD Palladum, Tamil Nadu.
[22] Aravindan C, Terrance Frederick Fernandez, Hema Malini V and
Catherine Madhu Vidha J, “An Extensive Research on Cyber Threats
using Learning Algorithm”, IEEE proceeding of International
Conference on Emerging Trends in Information Technology and
Engineering, ISBN: 978-1-7281-4141-1, 25th February 2020, Vellore,
India.