Content uploaded by Arnold Adimabua Ojugo
Author content
All content in this area was uploaded by Arnold Adimabua Ojugo on Oct 21, 2023
Content may be subject to copyright.
ISSN: 1001-0920
Volume 38, Issue 01, April, 2023
667
DeGATraMoNN: Deep Learning Memetic Ensemble to
Detect Spam Threats via a Content-Based Processing
Arnold Adimabua Ojugo1, Maureen Ifeanyi Akazue2, Patrick Ogholoruwami Ejeh3, Christopher
Chukwufunaya Odiakaose4, Frances Uche Emordi5
Department of Computer Science, Federal University of Petroleum Resources Effurun, Effurun, Nigeria1
Department of Computer Science, Delta State University Abraka, Delta State, Nigeria2
Department of Computer Science, Dennis Osadebey University, Asaba, Delta State, Nigeria3,4
Department of Cybersecurity, Dennis Osadebey University, Asaba, Delta State, Nigeria5
ABSTRACT— Technological growth is targeted at advancing society to higher sophistication with ease.
The birth and advances of the Internet to ease sharing of resources can be attributed to features such as its
ease of use, low transaction cost, ubiquitous nature, and user-trust level in adoption. These feats, while
advancing its popularity and adoption, also threaten data integrity with a rise in the birth of adversaries.
Thus, the quest of detecting intrusion and provisioning countermeasures is now a continuous task. Our study
advances a memetic deep-learning modular neural network to detect phishing attacks. Results show model
accurately detects malicious contents using a heuristic approach of text identification processing to detect
compromised links within an email. Our results show model performance with 98 percent sensitivity, an 87-
percent specificity, and 95 percent accuracy, respectively. The model also yields a 4-percent
misclassification error rate, for dataset acquisition alongside the generation of ruleset addition to the
knowledgebase that was not originally included and used for training of the model from the outset.
KEYWORDS: Credit-card, fraud detection, malware, phishing, spam and ham, transactions
1. INTRODUCTION
The Internet along with a plethora of connected enabled-device forms a giant network that connects over a
3.484billion users as of 2019. It advances an avenue to share resources among users, and is also rife as a
tool to exploit potential victims via phishing [1] – revealing how vulnerable, connected devices are within a
network. An attacker seeks to compromise a potential victim, and often succeeds based on the target's
judgment [2]; Rather, than on security measures in place on the victim's network. 2018 alone witnessed a
launch of fake Facebook pages with over 60% phishing attacks [3]. While in 2019, social media attacks saw
a 74.7-percent increase [4], [5]. These statistics affirm that the attacks are not receding; Rather, the thoughts
of the gains to be exploited therein – often provides for an attacker as entry point motivation against
potential compromises, and serve as a pivot point for attack propagation on a network [6- 9].
These scams and attacks have evolved to include network messages, emails, and SMS with over a 30%
increase as they have now migrated to social media platforms [10- 12]. They use spam (i.e. harmless
advertising) that is often laced with malware designed to exploit recipients. Spam attacks are deployed to
target high-volume, low-value victims [13], [14]. They require no coordinated expertise and are overlooked
as insignificant even with their estimated daily volume of over 422 billion in 2017 and 612 billion in 2020
to constitute about 85 percent of daily global traffic [15], [16] – with a distribution, eased by spammers who
are capable of sending tons of messages via botnets in seconds with recipients databank that are potentially
vulnerable due to ineffective anti-virus and other countermeasures [17].
Ojugo, et.al, 2023 KZYJC
668
Phishing is an identity theft aimed at stealing sensitive user data, from an unsuspecting, compromised
victim [18]. It may also involve the creation of compromised websites, acquisition of email lists and
botnets, and spoofing of emails/SMS targeting an unsuspecting victim to download its malicious contents as
originating from a trustworthy source [19]. A phishing attack has 3-elements: lure, hook, and catch. A lure
message is sent to a potential victim, and it seeks to exploit a potential victim’s desire for (a) curiosity to
click the compromised links, (b) fear that urges the potential victim to give out their data, and (c) empathy
as the attacker tries to impersonate a close associate in need [20], [21]. A hook message includes a
compromised link; while a catch obtains and uses extracted data. This technique constantly evolves to
reflect new trends and to evade detection [22]. The frequency and diversity of these attacks vary –
increasing their chances of success [9], [23- 25].
All scams are socially engineered to appeal to different human vulnerabilities – suggesting that some users
have ‘traits’ that make them susceptible to a scam [26], [27]. Some compromised users are scammed,
repeatedly. [8] some users are incapable of resisting persuasion and offers, which leaves them more
susceptible. Thus, over 20% of users are vulnerable to scams. Some may experience scams repeatedly [28].
Phishing success is more now, attributed to social engineering methods, which seek to appeal to a potential
victim’s emotions – as the attacker portends to create trust via personalized conversations. The attack
occurs in 2-stages: (a) first, an attacker befriends a potential victim [29] via a request that grants an attacker
access to a potential victim's details and contacts, and (b) the attacker then uses data as provided on the
(now) compromised victim’s page to requests for personalized data [30], or extends his attacks to contacts
on the victim’s page. Often, the attacker employs compromised links with malware attached therein that
will impact an unsuspecting victim’s device – as the phisher is often viewed as a mutual friend. Thus,
making believability easier and a bit more legitimate [31], [32].
2. Materials and Methods
2.1 Review of Related Literature
Studies have since begun to examine feats such as personality traits, demographics, web content, etc as
potential factors that contribute to victim susceptibility. [14] noted that people with high score neuroticism
showed a worse probability of detecting lies. They will rather believe friends are truthful to avoid the pains
of emotion. Premeditation is highly correlated with the capability to detect lies. Summarily, some studies
believe that some traits equip potential victims with an agreeable feat to detect lies; while others disagree
based on these features [33- 35]:
1. Personality Traits show a consistent pattern of how users respond to their environments and
different tasks. [36] investigated personality as 5-unique traits: (a) neuroticism as the tendency to
experience negative feelings (e.g. sadness and fear); And studies have shown that high neuroticism yields
increased susceptibility to irrational thoughts and less control over impulses, (b) conscientiousness is the
tendency to show high self-control and strong-willed, (c) openness is a willingness to try new
experiences, become more imaginative and intellectually curious, (d) agreeable is the willingness to help
others via believing, and (e) extroverted persons are more friendly and interact more – as supported by
[37- 39].
2. Demographics feats such as gender, online presence, and age as in previous studies have noted that
users ranging from 18-to-29years were more vulnerable to attacks based on email and web content; while
females between 24-to-42years were identified as most vulnerable due to their addiction and quest to
reduce social seclusion. Excess online presence and dependence create opportunities for phishers, which
in turn exposes a target’s close associates. Other studies have also linked age to risky behavior, and
increases their chances of being phished as young adults are less cautious about financial risk and; Thus,
ISSN: 1001-0920
Volume 38, Issue 01, April, 2023
669
less exposed to phishing training. Women are more vulnerable – as they are easier to entice via emails.
Other studies have also shown that women are equally capable and proficient as men in detecting a
deceptive message [40], [41].
3. Online Presence – A victim’s presence online can evolve into habits, leading to susceptibility.
Active users on social media are more susceptible. A user’s online habits influence how they process
attacks as they may be poised to click compromised links as well as respond to conversations without
engaging in adequate cognition and/or granting adequate attention to their perceived online behavior [14]
– leading to their being lured, hooked and captured over social media sites. It increases the chances of a
user thoughtlessly clicking a malicious link, or accepting a request from a faked profile without re-course
to the impact of such an action. Actions like clicking on links, and sharing/liking wall posts without
recourse to details on web content – can often impact negatively the user’s trust and risk perception [42],
[43].
4. Media Content Processing can be achieved via 2-modes: heuristic and systemic. Systemic
processing allows a user to perceive if the available data is sufficient to determine a choice to be
undertaken. Most user utilizes both heuristics (i.e. take cues that impact the limited cognitive resources to
judge and decide), and systemic (i.e. involves a careful examination of the data to reach a decision) [44],
[45]. The risk of message overload on websites encourages users to heuristically, process content. This
often leads to quick and effortless judgment [46], [47]. While heuristic processing is more efficient, it
significantly increases attack vulnerability – as users will then overlook cues that suggest a message is
malicious and poses a threat. Heuristic content processing relies on the credibility of cues and lures that
leads users into trusting phishing conversation cum messages [48].
2.2 Motivations / Statement of Problem
This study is motivated by the following problems:
1. Detection of phishing spam is not often reported. And, when and if reported – they are often limited
in their implementation strategies.
2. It is quite unwise for the detection employed by organizations for the detection of phishing to be de-
scribed in great detail over such an uncensored public domain – as such will further equip phishers and
attackers with the requisite knowledge required to evade detection.
3. Limited data availability and censored results often make implementing these schemes, tedious to
de-velop and deploy
4. Many of these schemes are also rippled with unreliable performance, which often results from noisy
data, improper cum mismatch of features/parameters selected for use, and anomalies. However, noisy
features can be eliminated by accurately optimizing our classifier.
Some users have been noted to remain online considerably, engage in repetitive actions that form, sug-gest,
and help establish habit patterns such as liking pages and posts, and not thoroughly analyze conversa-tions
or comments adequately. To curb such, we propose a deep-learning memetic network to detect mali-cious
content via a hybrid heuristic–systemic paradigm such that when reviewing communication, partici-pants
can engage in two modes of content processing.
2.3 Proposed Experimental Phishing Detection Ensemble
We use a hybrid deep learning model as seen in figure 1, which consist of 3-blocks as adapted by [49], [50]
as follows: (a) deep learning modular Kohonen neural net-work, (b) the supervised cultural genetic
algorithm, and (c) the knowledgebase.
Ojugo, et.al, 2023 KZYJC
670
Figure 1. Phishing Detection Framework / Architecture
1. The Cultural Genetic Algorithm: Basically, a GA-block uses 4 operators (initialize, fitness function
and select, mutation, and crossover) to uncover probable solution(s). A gene is fit – if its value is close to
optimal. A variant of GA is the Cultural GA (CGA), which uses 4-belief spaces to define its solution
space namely: (a) the normative belief which defines the specific value ranges to which a gene is bound,
(b) domain belief contains knowledge about the task being undertaken, (c) temporal belief con-tains
knowledge about the available problem space, and (d) spatial belief contains knowledge about the
topography for the task. Furthermore, it uses the influence function to bridge the belief spaces and the
gene pool – to ensure any modified genes still conform to the belief space(s). The CGA should yield a
result pool that does not violate its belief space and assist in reducing the number of potential genes
generated by the GA until an optimum is discovered [13], [51].
2. The Kohonen Modular Network (MNN) is a gridlike, feed-forward network whose first layer
accepts input, and re-sends unbound to its second layer, which uses the transfer function to offer
competitive computation. The competitive layer then maps similarity patterns into relations. Pattern
relations noticed are used to determine the result after training [49], [52]. We modify the parameters and
carefully create our deep-learning Kohonen MNN through a deep architecture. Our deep learning is
achieved by training the network component via 2-stages namely the pre-trained, and fine-tuned processes
as described in [53], [54].
So, we adapt [55], [56], [50] to our experimental ensemble (see figure 2) and are trained as thus [57]:
Figure 2. Deep Learning Genetic Algorithm Trained Modular Neural Network (DeGATraMoNN)
1. Input: The dataset forms the input used to train the GA-unit with operators as encoder, selector,
Training / Testing
Spam Data
Contents
Knowledge
Base
Feature Extraction
Lexical Patterns
URL
String-Matching
HTML
Deep Learning
Initialization
Feat Optimization
Trained Classifier
Model Evaluation
Decision Support
Result(s)
Confusion Matrix
Error Analysis
ISSN: 1001-0920
Volume 38, Issue 01, April, 2023
671
swap-per, changer, and terminator. The knowledgebase stores the optimized and learned dataset features
as classified.
2. Our Deep MNN receives the optimized data as labeled and unlabeled rules – and propagated it as
en-hanced, predefined data (i.e. genuine- and benign- classes) via the network. Each rule is a systemic
pro-duction structured to consist of 4-modules: (i) a rule set, where each rule contains how it is to be ap-
plied, and the corresponding task to be performed or carried out by the rule, (b) a knowledgebase of op-
timized rules stored in classes, (c) control strategy of how the rules are searched within a knowledge-base
to successfully find a match, and the how to resolve conflicts if several rules are matched at the same
time, and (d) MNN applier also functions as the principal component analyzer with rules im-proved using
GA's diversity and recombination so that a trained network can autonomously and suc-cessfully classify
transactions onto both classes.
3. The decision support of predicted rule output automatically updates the ensemble’s knowledge
base. And where there are new data acquired, corresponding transaction rules are also generated and now
en-countered within the model. And are thus, subsequently classified as either normal data type or mali-
cious data type.
2.4 Parameters/Features Tuning and Estimation
The study adopts the KDD’99 CUPS dataset. The dataset is split into training (70%) and testing (30%). The
rule-based optimized dataset's data labels are used to identify a model's prediction ability. At the input layer
of our deep learning Kohonen map, we use 10 neurons (a neuron for each feature). The out-put layer is
made up of two neurons (a neuron for each possible class of normal and benign rules). The learning rate(s),
epoch size, transfer function, and hidden layer structure, are among the parameters to be tuned. Thus, we
used a 500-epoch Rectified Linear Unit Transfer Function [58]. Mindful of our model’s mean convergence
time and precision, optimal values were found with epoch configuration (of 100, 300, and 500 respectively)
[59], [60] to yield the least amount of error, and best-fit results. The trial-and-error method was used to
determine the number of hidden layers. For rules classified by normal and harmful content classes, the
model generated 22-fit rules partitioned to correspond to an array of chromosomes [61], [62].
3. Findings and Discussion
3.1 Result and Findings
Simulation on testbeds with a single-layered network of 1-to-10 neurons yields the highest f-score and least
training loss time to result in the best number of layers. Adding a second and a third hidden layer also
resulted yielded good results with the highest number of neurons yielding the best scores. Table 1 shows the
resulting analysis in the use of the first hidden layers.
Table 1. First hidden layer configuration analysis
Hidden Layer
Precision
Recall
F1 Score
Iteration
Train Loss
Epoch
1
0.88
0.91
0.89
41
0.249
500
2
0.81
0.93
0.85
29
0.287
500
3
0.86
0.91
0.89
28
0.392
500
4
0.80
0.90
0.90
19
0.051
500
5
0.87
0.58
0.77
21
1.469
500
6
0.91
0.92
0.91
18
1.440
500
7
0.88
0.59
0.74
18
2.023
500
8
0.94
0.85
0.88
9
2.273
500
9
0.93
0.94
0.94
17
1.140
500
10
0.94
0.91
0.92
20
1.797
500
11
0.86
0.92
0.87
27
2.413
500
Ojugo, et.al, 2023 KZYJC
672
12
0.90
0.93
0.88
18
2.232
500
For the overall (best) with only one-hidden layer configuration, we have that the 9th iteration yields the best
overall result. It has 14 neurons with a 92% f-score, and 1.140 training loss. An f-score shows each run’s
accuracy within an unbalanced dataset used to train/test the model. Table 2 shows the first layer
configuration with 10 neurons and extra 2 neurons for optimal extra processing. The hidden layers of 9,11-
neurons resulted in a 93% f-score and 0.39 training loss value. The ensemble favors the adoption and con-
sequent use of a second hidden layer with a greater value for the f-score as in agreement with [63- 66].
Table 2. Second hidden layer configuration analysis
Hidden Layer
Precision
Recall
F1 Score
Iteration
Training Loss
Epoch
9, 1
0.91
0.92
0.83
29
0.393
500
9, 2
0.93
0.92
0.85
24
0.392
500
9, 3
0.91
0.92
0.90
25
0.483
500
9, 4
0.90
0.87
0.89
25
1.185
500
9, 5
0.58
0.92
0.91
18
1.482
500
9, 6
0.92
0.92
0.86
19
1.699
500
9, 7
0.59
0.92
0.89
22
0.318
500
9, 8
0.85
0.93
0.90
14
1.484
500
9, 9
0.94
0.92
0.91
19
1.659
500
9, 10
0.91
0.92
0.92
18
1.371
500
9, 11
0.92
0.94
0.93
14
0.390
500
9, 12
0.93
0.93
0.94
16
1.280
500
3.2 Discussion of Findings
To evaluate performance, we adapt classification and improvement percentage rates (as compared for both
the train/test datasets) as provisioned by Eq. 4 – to yield the summary of obtained values for classification
error rates as seen in Table 3; While, Equation 5 is the improvement rate of the ensemble equation which
yields the improvement rate as seen in Table 4.
Table 3. Classification Rate of Each model
Model/Frameworks
Classification Errors
Training
Testing
Experimental Ensemble
1.29%
1.09%
Benchmark Models
GANN
21.3%
19.7%
PHMM
13.7%
10.2%
Table 4. Improvement Percentage
Model/Framework
Training
Testing
Proposed Ensemble
75.89%
92.01%
Benchmark Models
GANN
42.79%
34.09%
PHMM
56.03%
64.16%
ISSN: 1001-0920
Volume 38, Issue 01, April, 2023
673
Tables 3 and 4 respectively – shows that PHMM outperforms GANN with a 13.7% classification error rate
(i.e. false positives and true negatives). It yields 87.3% classification accuracy with a 56.03% improvement
rate at training. Conversely, GANN has a misclassification rate of 21.3% (false-positives and true-negatives
error rate) and promises an improvement of 42.79%. Both underperformed when compared to our proposed
hybrid Genetic Algorithm trained Modular neural network as in tables 3 and 4 respectively. It follows thus
to reason why and how many social media platforms including Facebook, can then regularly retrieve and
analyze web content – to investigate, blacklist, and prosecute phishers [21], [23], [67], [68].
3.3 Model Evaluation and Performance
To compute the sensitivity, specificity, and accuracy of the ensemble – we evaluate its performance using
Eq. 1 to Eq. 3 respectively. So, given that TN = 7, FP = 1, TP = 53, and FN = 1. We then compute thus:
The ensemble is found to yield a sensitivity of 98%, specificity of 87.5%, and prediction accuracy of 95%
for data not included from outset (i.e. not originally used to train the model).
4. Conclusion
It is worth noting that social media networks have rules and safeguards in place to educate consumers and
protect them against phishing. For example, Facebook’s phish@fb.com is a dedicated email address for
reporting phishing attempts. This gives Facebook the capability to investigate, blacklist, and prosecute
phishers. They also provide information and instructions for users who have been phished on Facebook or
whose device has been infected with malware. Both the social media platform and the user – are held
accountable for phishing attack prevention, dissolution, reporting, and awareness. The social media
platform is in charge of informing users about phishing and giving controls to prevent them. Conversely,
users must stay ahead of the curve with and about preventing these attacks as well as implementing safety
controls to limit such accidents.
5. References
[1] S. Goel, K. Williams, and E. Dincelli, “Got Phished? Internet Security and Human Vulnerability,”
J. Assoc. Inf. Syst., vol. 18, no. 1, pp. 22–44, Jan. 2017, doi: 10.17705/1jais.00447.
[2] A. A. Ojugo and R. E. Yoro, “Forging a deep learning neural network intrusion detection
framework to curb the distributed denial of service attack,” Int. J. Electr. Comput. Eng., vol. 11, no. 2, pp.
1498–1509, 2021, doi: 10.11591/ijece.v11i2.pp1498-1509.
[3] R. Sobers, “166 Cybersecurity statistics and trends,” Data Secur., vol. 12, pp. 23–29, 2022.
[4] I. Benchaji, S. Douzi, B. El Ouahidi, and J. Jaafari, “Enhanced credit card fraud detection based on
attention mechanism and LSTM deep model,” J. Big Data, vol. 8, no. 1, p. 151, 2021, doi: 10.1186/s40537-
021-00541-8.
Ojugo, et.al, 2023 KZYJC
674
[5] C. Li, N. Ding, H. Dong, and Y. Zhai, “Application of Credit Card Fraud Detection Based on CS-
SVM,” Int. J. Mach. Learn. Comput., vol. 11, no. 1, pp. 34–39, 2021, doi: 10.18178/ijmlc.2021.11.1.1011.
[6] A. Jayatilaka, N. A. G. Arachchilage, and M. A. Babar, “Falling for Phishing: An Empirical
Investigation into People’s Email Response Behaviors,” arXiv Prepr. arXiv …, no. Fbi 2020, pp. 1–17,
2021.
[7] D. Huang, Y. Lin, Z. Weng, and J. Xiong, “Decision Analysis and Prediction Based on Credit Card
Fraud Data,” in The 2nd European Symposium on Computer and Communications, Apr. 2021, pp. 20–26.
doi: 10.1145/3478301.3478305.
[8] I. A. Anderson and W. Wood, “Habits and the electronic herd: The psychology behind social
media’s successes and failures,” Consum. Psychol. Rev., vol. 4, no. 1, pp. 83–99, Jan. 2021, doi:
10.1002/arcp.1063.
[9] Y. Lucas et al., “Multiple perspectives HMM-based feature engineering for credit card fraud
detection,” in Proceedings of the 34th ACM/SIGAPP Symposium on Applied Computing, Apr. 2019, pp.
1359–1361. doi: 10.1145/3297280.3297586.
[10] N. Rtayli and N. Enneya, “Enhanced credit card fraud detection based on SVM-recursive feature
elimination and hyper-parameters optimization,” J. Inf. Secur. Appl., vol. 55, p. 102596, Dec. 2020, doi:
10.1016/j.jisa.2020.102596.
[11] H. Tingfei, C. Guangquan, and H. Kuihua, “Using Variational Auto Encoding in Credit Card Fraud
Detection,” IEEE Access, vol. 8, pp. 149841–149853, 2020, doi: 10.1109/ACCESS.2020.3015600.
[12] M. I. Akazue, R. E. Yoro, B. O. Malasowe, O. Nwankwo, and A. A. Ojugo, “Improved services
traceability and management of a food value chain using block-chain network : a case of Nigeria,” Indones.
J. Electr. Eng. Comput. Sci., vol. 29, no. 3, pp. 1623–1633, 2023, doi: 10.11591/ijeecs.v29.i3.pp1623-1633.
[13] S. M. Albladi and G. R. S. Weir, “User characteristics that influence judgment of social engineering
attacks in social networks” Human-centric Comput. Inf. Sci., vol. 8, no. 1, 5, Dec. 2018, doi:
10.1186/s13673-018-0128-7.
[14] C. L. Rash and S. M. Gainsbury, “Disconnect between intentions and outcomes: A comparison of
regretted text and photo social networking site posts,” Hum. Behav. Emerg. Technol., vol. 1, no. 3, pp. 229–
239, Jul. 2019, doi: 10.1002/hbe2.165.
[15] D. Zhang, B. Bhandari, and D. Black, “Credit Card Fraud Detection Using Weighted Support
Vector Machine,” Appl. Math., vol. 11, no. 12, pp. 1275–1291, 2020, doi: 10.4236/am.2020.1112087.
[16] E. R. Altman, “Synthesizing Credit Card Transactions,” Oct. 2019, [Online]. Available:
http://arxiv.org/abs/1910.03033
[17] M. Gratian, S. Bandi, M. Cukier, J. Dykstra, and A. Ginther, “Correlating human traits and cyber
security behavior intentions,” Comput. Secur., vol. 73, pp. 345–358, Mar. 2018, doi:
10.1016/j.cose.2017.11.015.
ISSN: 1001-0920
Volume 38, Issue 01, April, 2023
675
[18] A. A. Ojugo and R. E. Yoro, “Extending the three-tier constructivist learning model for alternative
delivery: ahead the COVID-19 pandemic in Nigeria,” Indones. J. Electr. Eng. Comput. Sci., vol. 21, no. 3,
p. 1673, Mar. 2021, doi: 10.11591/ijeecs.v21.i3.pp1673-1682.
[19] A. A. Ojugo, C. O. Obruche, and A. O. Eboka, “Quest For Convergence Solution Using Hybrid
Genetic Algorithm Trained Neural Network Model For Metamorphic Malware Detection,” ARRUS J. Eng.
Technol., vol. 2, no. 1, pp. 12–23, Nov. 2021, doi: 10.35877/jetech613.
[20] W. Rocha Flores, H. Holm, M. Nohlberg, and M. Ekstedt, “Investigating personal determinants of
phishing and effect of national culture,” Inf. Comput. Secur., vol. 23, no. 2, pp. 178–199, 2015, doi:
10.1108/ICS-05-2014-0029.
[21] M. Barlaud, A. Chambolle, and J.-B. Caillau, “Robust supervised classification and feature
selection using a primal-dual method,” Feb. 2019.
[22] A. A. Ojugo and O. D. Otakore, “Intelligent cluster connectionist recommender system using
implicit graph friendship algorithm for social networks,” IAES Int. J. Artif. Intell., vol. 9, no. 3, p. 497~506,
2020, doi: 10.11591/ijai.v9.i3.pp497-506.
[23] A. A. Ojugo and D. A. Oyemade, “Boyer moore string-match framework for a hybrid short
message service spam filtering technique,” IAES Int. J. Artif. Intell., vol. 10, no. 3, pp. 519–527, 2021, doi:
10.11591/ijai.v10.i3.pp519-527.
[24] G. Sasikala et al., “An Innovative Sensing Machine Learning Technique to Detect Credit Card
Frauds in Wireless Communications,” Wirel. Commun. Mob. Comput., vol. 2022, pp. 1–12, Jun. 2022, doi:
10.1155/2022/2439205.
[25] M. Laavanya and V. Vijayaraghavan, “Real Time Fake Currency Note Detection using Deep
Learning,” Int. J. Eng. Adv. Technol., vol. 9, no. 1S5, pp. 95–98, 2019, doi:
10.35940/ijeat.a1007.1291s52019.
[26] P. Moodley, D. C. S. Rorke, and E. B. Gueguim Kana, “Development of artificial neural network
tools for predicting sugar yields from inorganic salt-based pretreatment of lignocellulosic biomass,”
Bioresour. Technol., vol. 273, pp. 682–686, Feb. 2019, doi: 10.1016/j.biortech.2018.11.034.
[27] D. Wang, B. Chen, and J. Chen, “Credit card fraud detection strategies with consumer incentives,”
Omega, vol. 88, pp. 179–195, Oct. 2019, doi: 10.1016/j.omega.2018.07.001.
[28] M. Al-Qatf, Y. Lasheng, M. Al-Habib, and K. Al-Sabahi, “Deep Learning Approach Combining
Sparse Autoencoder With SVM for Network Intrusion Detection,” IEEE Access, vol. 6, pp. 52843–52856,
2018, doi: 10.1109/ACCESS.2018.2869577.
[29] E. O. Yeboah-Boateng and P. M. Amanor, “Phishing , SMiShing & Vishing : An Assessment of
Threats against Mobile Devices,” J. Emerg. Trends Comput. Inf. Sci., vol. 5, no. 4, pp. 297–307, 2014.
[30] A. A. Ojugo and A. O. Eboka, “Signature-Based Malware Detection Using Approximate Boyer
Moore String Matching Algorithm,” Int. J. Math. Sci. Comput., vol. 5, no. 3, pp. 49–62, 2019, doi:
Ojugo, et.al, 2023 KZYJC
676
10.5815/ijmsc.2019.03.05.
[31] A. Algarni, Y. Xu, and T. Chan, “An empirical study on the susceptibility to social engineering in
social networking sites: the case of Facebook,” Eur. J. Inf. Syst., vol. 26, no. 6, pp. 661–687, Nov. 2017,
doi: 10.1057/s41303-017-0057-y.
[32] Y. Zhang, S. Egelman, L. F. Cranor, and J. Hong, “Phinding Phish: Evaluating Anti-Phishing
Tools,” Proc. Netw. Distrib. Syst. Secur. Symp. (NDSS 2007), no. March, pp. 1–16, 2007.
[33] R. E. Yoro, F. O. Aghware, B. O. Malasowe, O. Nwankwo, and A. A. Ojugo, “Assessing
contributor features to phishing susceptibility amongst students of petroleum resources varsity in Nigeria,”
Int. J. Electr. Comput. Eng., vol. 13, no. 2, pp. 1922–1931, 2023, doi: 10.11591/ijece.v13i2.pp1922-1931.
[34] R. E. Yoro, F. O. Aghware, M. I. Akazue, A. E. Ibor, and A. A. Ojugo, “Evidence of personality
traits on phishing attack menace among selected university undergraduates in Nigerian,” Int. J. Electr.
Comput. Eng., vol. 13, no. 2, pp. 1943–1953, Apr. 2023, doi: 10.11591/ijece.v13i2.pp1943-1953.
[35] M. I. Akazue, A. A. Ojugo, R. E. Yoro, B. O. Malasowe, and O. Nwankwo, “Empirical evidence of
phishing menace among undergraduate smartphone users in selected universities in Nigeria,” Indones. J.
Electr. Eng. Comput. Sci., vol. 28, no. 3, pp. 1756–1765, Dec. 2022, doi: 10.11591/ijeecs.v28.i3.pp1756-
1765.
[36] H. Yildiz Durak, “Human Factors and Cybersecurity in Online Game Addiction: An Analysis of the
Relationship Between High School Students’ Online Game Addiction and the State of Providing Personal
Cybersecurity and Representing Cyber Human Values in Online Games,” Soc. Sci. Q., vol. 100, no. 6, pp.
1984–1998, Oct. 2019, doi: 10.1111/ssqu.12693.
[37] T. Halevi, J. Lewis, and N. Memon, “A pilot study of cyber security and privacy related behavior
and personality traits,” in Proceedings of the 22nd International Conference on World Wide Web, May
2013, pp. 737–744. doi: 10.1145/2487788.2488034.
[38] J. Camargo and A. Young, “Feature Selection and Non-Linear Classifiers: Effects on Simultaneous
Motion Recognition in Upper Limb,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 27, no. 4, pp. 743–750,
Apr. 2019, doi: 10.1109/TNSRE.2019.2903986.
[39] Maya Gopal P S and Bhargavi R, “Selection of Important Features for Optimizing Crop Yield
Prediction,” Int. J. Agric. Environ. Inf. Syst., vol. 10, no. 3, pp. 54–71, Jul. 2019, doi:
10.4018/IJAEIS.2019070104.
[40] L. De Kimpe, M. Walrave, W. Hardyns, L. Pauwels, and K. Ponnet, “You’ve got mail! Explaining
individual differences in becoming a phishing target,” Telemat. Informatics, vol. 35, no. 5, pp. 1277–1287,
Aug. 2018, doi: 10.1016/j.tele.2018.02.009.
[41] A. Vishwanath, “Habitual Facebook Use and its Impact on Getting Deceived on Social Media,” J.
Comput. Commun., vol. 20, no. 1, pp. 83–98, Jan. 2015, doi: 10.1111/jcc4.12100.
[42] K. Parsons, A. McCormac, M. Pattinson, M. Butavicius, and C. Jerram, “The design of phishing
ISSN: 1001-0920
Volume 38, Issue 01, April, 2023
677
studies: Challenges for researchers,” Comput. Secur., vol. 52, pp. 194–206, Jul. 2015, doi:
10.1016/j.cose.2015.02.008.
[43] P. Kumaraguru, S. Sheng, A. Acquisti, L. F. Cranor, and J. Hong, “Teaching Johnny not to fall for
phish,” ACM Trans. Internet Technol., vol. 10, no. 2, pp. 1–31, May 2010, doi: 10.1145/1754393.1754396.
[44] V. Pandiyaraju, R. Logambigai, S. Ganapathy, and A. Kannan, “An Energy Efficient Routing
Algorithm for WSNs Using Intelligent Fuzzy Rules in Precision Agriculture,” Wirel. Pers. Commun., vol.
112, no. 1, pp. 243–259, May 2020, doi: 10.1007/s11277-020-07024-8.
[45] S. Verma, A. Bhatia, A. Chug, and A. P. Singh, “Recent Advancements in Multimedia Big Data
Computing for IoT Applications in Precision Agriculture: Opportunities, Issues, and Challenges,” 2020, pp.
391–416. doi: 10.1007/978-981-13-8759-3_15.
[46] P. Charoen-Ung and P. Mittrapiyanuruk, “Sugarcane Yield Grade Prediction Using Random Forest
with Forward Feature Selection and Hyper-parameter Tuning,” 2019, pp. 33–42. doi: 10.1007/978-3-319-
93692-5_4.
[47] S. Khaki, L. Wang, and S. V. Archontoulis, “A CNN-RNN Framework for Crop Yield Prediction,”
Front. Plant Sci., vol. 10, Jan. 2020, doi: 10.3389/fpls.2019.01750.
[48] H. J. Parker and S. V. Flowerday, “Contributing factors to increased susceptibility to social media
phishing attacks,” SA J. Inf. Manag., vol. 22, no. 1, Jun. 2020, doi: 10.4102/sajim.v22i1.1176.
[49] G. Behboud, “Reasoning using Modular Neural Network,” Towar. Data Sci., vol. 34, no. 2, pp. 12–
34, 2020.
[50] A. A. Ojugo and O. Nwankwo, “Spectral-Cluster Solution For Credit-Card Fraud Detection Using
A Genetic Algorithm Trained Modular Deep Learning Neural Network,” JINAV J. Inf. Vis., vol. 2, no. 1,
pp. 15–24, Jan. 2021, doi: 10.35877/454RI.jinav274.
[51] X. Lin, P. R. Spence, and K. A. Lachlan, “Social media and credibility indicators: The effect of
influence cues,” Comput. Human Behav., vol. 63, pp. 264–271, Oct. 2016, doi: 10.1016/j.chb.2016.05.002.
[52] L. A. Belanche and F. F. González, “Review and Evaluation of Feature Selection Algorithms in
Synthetic Problems,” Inf. Fusion, vol. 23, pp. 34–54, Jan. 2011.
[53] Y. Abakarim, M. Lahby, and A. Attioui, “An Efficient Real Time Model For Credit Card Fraud
Detection Based On Deep Learning,” in Proceedings of the 12th International Conference on Intelligent
Systems: Theories and Applications, Oct. 2018, pp. 1–7. doi: 10.1145/3289402.3289530.
[54] R. J. Urbanowicz, M. Meeker, W. La Cava, R. S. Olson, and J. H. Moore, “Relief-based feature
selection: Introduction and review,” J. Biomed. Inform., vol. 85, pp. 189–203, Sep. 2018, doi:
10.1016/j.jbi.2018.07.014.
[55] M. Dadkhah, T. Sutikno, J. M. Davarpanah, and D. Stiawan, “An Introduction to Journal Phishings
and Their Detection Approach,” TELKOMNIKA, vol. 13, no. 2, p. 373, Jun. 2015, doi:
Ojugo, et.al, 2023 KZYJC
678
10.12928/telkomnika.v13i2.1436.
[56] O. V. Lee et al., “A malicious URLs detection system using optimization and machine learning classifiers,”
Indones. J. Electr. Eng. Comput. Sci., vol. 17, no. 3, p. 1210, 2020, doi: 10.11591/ijeecs.v17.i3.pp1210-1214.
[57] A. A. Ojugo and A. O. Eboka, “Memetic algorithm for short messaging service spam filter using text
normalization and semantic approach,” Int. J. Informatics Commun. Technol., vol. 9, no. 1, p. 9, 2020, doi:
10.11591/ijict.v9i1.pp9-18.
[58] Y. Ampatzidis, V. Partel, and L. Costa, “Agroview: Cloud-based application to process, analyze and visualize
UAV-collected data for precision agriculture applications utilizing artificial intelligence,” Comput. Electron. Agric.,
vol. 174, p. 105457, Jul. 2020, doi: 10.1016/j.compag.2020.105457.
[59] X. E. Pantazi, D. Moshou, T. Alexandridis, R. L. Whetton, and A. M. Mouazen, “Wheat yield prediction
using machine learning and advanced sensing techniques,” Comput. Electron. Agric., vol. 121, pp. 57–65, Feb. 2016,
doi: 10.1016/j.compag.2015.11.018.
[60] A. Goldstein, L. Fink, A. Meitin, S. Bohadana, O. Lutenberg, and G. Ravid, “Applying machine learning on
sensor data for irrigation recommendations: revealing the agronomist’s tacit knowledge,” Precis. Agric., vol. 19, no. 3,
pp. 421–444, Jun. 2018, doi: 10.1007/s11119-017-9527-4.
[61] D. Nahavandi, R. Alizadehsani, A. Khosravi, and U. R. Acharya, “Application of artificial intelligence in
wearable devices: Opportunities and challenges,” Comput. Methods Programs Biomed., vol. 213, no. December, p.
106541, Jan. 2022, doi: 10.1016/j.cmpb.2021.106541.
[62] D. Liu and W. K. Campbell, “The Big Five personality traits, Big Two metatraits and social media: A meta-
analysis,” J. Res. Pers., vol. 70, pp. 229–240, Oct. 2017, doi: 10.1016/j.jrp.2017.08.004.
[63] M. Zareapoor and P. Shamsolmoali, “Application of Credit Card Fraud Detection: Based on Bagging
Ensemble Classifier,” Procedia Comput. Sci., vol. 48, pp. 679–685, 2015, doi: 10.1016/j.procs.2015.04.201.
[64] Y. Gao, S. Zhang, J. Lu, Y. Gao, S. Zhang, and J. Lu, “Machine Learning for Credit Card Fraud Detection,”
in Proceedings of the 2021 International Conference on Control and Intelligent Robotics, Jun. 2021, pp. 213–219. doi:
10.1145/3473714.3473749.
[65] P. . Maya Gopal and Bhargavi R, “Feature Selection for Yield Prediction Using BORUTA Algorithm,” Int. J.
Pure Appl. Math., vol. 118, no. 22, pp. 139–144, 2018.
[66] S. Yuan and X. Wu, “Deep learning for insider threat detection: Review, challenges and opportunities,”
Comput. Secur., vol. 104, 2021, doi: 10.1016/j.cose.2021.102221.
[67] R. Broadhurst, K. Skinner, N. Sifniotis, and B. Matamoros-Macias, “Cybercrime Risks in a University
Student Community,” SSRN Electron. J., no. May, 2018, doi: 10.2139/ssrn.3176319.
[68] A. A. Ojugo, A. O. Eboka, R. E. Yoro, M. O. Yerokun, and F. N. Efozia, “Hybrid model for early
diabetes diagnosis,” Math. Comput. Ind., vol. 50, no. 3–5, pp. 55–65, 2015, doi: 10.1109/MCSI.2015.35.
This work is licensed under a Creative Commons Attribution Non-Commercial 4.0
International License.