Proceedings of the Third International Conference on Smart Systems and Inventive Technology (ICSSIT 2020)

IEEE Xplore Part Number: CFP20P17-ART; ISBN: 978-1-7281-5821-1

Phishing Attacks Detection using Machine

Learning Approach
M ohammad Nazmul Alam Dhiman Sarma Farzana Firoz Lima
Department of Computer Science and Department of Computer Science and Department of Computer Science and
Engineering Engineering Engineering
Royal University of Dhaka Rangamati Science and Technology East Delta University
Dhaka, Bangladesh University Chittagong, Bangladesh
[email protected] Rangamati, Bangladesh [email protected]
[email protected]

Ishita Saha Rubaiath-E- Ulfath Sohrab Hossain

Department of Computer Science and Department of Computer Science and Department of Computer Science and
Engineering Engineering Engineering
University of Science and Technology East Delta University East Delta University
Chittagong Chittagong , Bangladesh Chittagong, Bangladesh
[email protected] [email protected] [email protected]

Abstract- Evolving digital transformation has exacerbated account information, and credit card details for the
cybersecurity threats globally. Digitization expands the doors transaction. Fishers always change their strategy to attack
wider to cybercriminals. Initially cyber threats approach in the the system. Social engineering [3-6] is one of the essential
form of phishing to steal the confidential user credentials.
techniques the fishers use. Using this technique, they gather
Usually, Hackers will influence the users through phishing in
personal credentials from a trustworthy person. Phishers
order to gain access to the organization's digital assets and
networks. With security breaches, cybercriminals execute create false websites and spoof email in such a way that they
ransomware attack, get unauthorized access, and shut down are very similar and sometimes look like a real company
systems and even demand a ransom for releasing the access. website that comes from a source. Sometimes the attackers
Anti-phishing software and techniques are circumvented by act like a real source and force the users to update the
the phishers for dodging tactics. Though threat intelligence and system.
behavioural analytics systems support organizations to spot the
unusual traffic patterns, still the best practice to prevent Moreover, they threaten the customer to suspend the
phishing attacks is defended in depth. In this perspective, the account and demand ransom. Email spoofing is another
proposed research work has developed a model to detect the technique used for phishing fraud [7]. Customers are usually
phishing attacks using machine learning (ML) algorithms like
misled to disclose private information like passwords and
random forest (RF) and decision tree (DT). A standard
credit card number. Thus fishing is mainly used to steal
legitimate dataset of phishing attacks from Kaggle was aided
for ML processing. To analyze the attributes of the dataset, the
valuable information such as bank account, password , and
proposed model has used feature selection algorithms like credit card details [8]. This type of scam is increasing
principal component analysis (PCA). Finally, a maximum rapidly, and individuals, business -people are losing their
accuracy of 97% was achieved through the random forest trust in online business. Thus, a negative impression of
algorithm. clients on online business was swarmed as they lost faith in
online transactions. Even though encryption software is used
Keywords—Phishing attack; phishing attack detection; artificial
to protect the information in the computers' storage, they are
intelligence; machine learning; deep learning; convolutional
also vulnerable to attacks [9]. In this paper, the detection of
neural network
fishing was performed through ML.
I. INT RODUCT ION
II. BACKGROUND AND RELAT ED W ORK
Phishing attacks have become anxiety for the cyber world. It
There are various types of phishing attacks used to cheat the
causes enormous problems for privacy and financial issues
users. Besides, various phishing detection techniques and
of internet users. Scammers, namely fishers, create false
tools are also available to defend phishing attacks .
websites [1, 2] to feel and look like a genuine to deceive the
Classification is one of the techniques used to detect website
people. They spoof emails to steal the identity of legitimate
users. They gather personal covert information, password,

978-1-7281-5821-1/20/$31.00 ©2020 IEEE 1173

Authorized licensed use limited to: BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE. Downloaded on December 13,2021 at 12:36:58 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Smart Systems and Inventive Technology (ICSSIT 2020)
IEEE Xplore Part Number: CFP20P17-ART; ISBN: 978-1-7281-5821-1

phishing [10, 11, 29]. Here, common types of a phishing party with attachment or link. They request to send an
attack and classification techniques are described below: updated version of the original [13].

A. Different types of phishing attacks 7) Whaling

Attackers use various methods to explore the vulnerability
This type of phishing is targeted at the higher executives of
of internet security. They always try to discover the
the organization. The content of the email is about important
loopholes in the security system to exploit them. Various
issues, and it is sent to the executives. The message of the
phishing assaults that are distinctive from each other are
email can be like the customers ' complaints.
presented below.
8) Spear phishing
1) Algorithm-Based Phishing
Spear phishing is one kind of email scam used to target
The first phishing attack was caught by America Online specific persons and organizations. The attacker sends
(AOL), which was created using an algorithm. The cheater emails to get a reply from the targeted person. The email in
built an algorithm to match the credit card number of such a way that they pretend to know many things about the
America Online accounts [3]. victims, such as their name, work address, email address,
and so on [14, 15].
2) Deceptive Phishing
B. Classification techniques to detect website phishing
Deceiver uses sorts of ways to deceit the internet users.
There are various techniques to fight against phishing
Fishers send emails to the users to verify the account. They
attacks to provide safety measures of internet users [16-18].
use links and buttons and request to click them. Behind the
Among all types of attacks, email spoofing and URL
links, there has a website where the hackers snatch and store
phishing are hard to identify and to stop because the attacker
the personal information of the users.
always changes their strategies. To stop these types of
3) URL Phishing phishing, it was recommended to block malicious emails
and bogus URLs [2].
Universal Resource Locator (URL) phishing is another kind
of phishing attack done by using a hidden link. The link A proactive strategy was taken to identify malicious URLs
holds the hackers' website. When the user clicks the link, it using lexical analysis [19]. To categories the untruthful
is redirected to the hackers' website and stores the user's URLs, machine learning algorithms were applied to
information [11]. features. The analysis was carried out on these features set
to mitigate these URLs. Furthermore, principal component
4) Hosts File Poisoning analysis (PCA) technique and random forest algorithm were
used for phishing attack detection. PCA was used to identify
It is used to poison the host file in the windows operating
principal components from the variables. After that, data
system platform. When the user locates the desired website,
were analyzed to detect phishing type. Random Forest
then it rerouted to hackers' site, or sometimes it replies 'The
algorithm was applied to the datasets. It classified the
Page Not Found'. If it can redirect to the fake site, then users
website and detected phishing. This method has high
data are recorded and stolen.
accuracy. It can classify the phishing email about 96 per
5) Content-Injection Phishing cent and easy to construct. Moreover, it can handle large
volumes of datasets. Labelled datasets were used for
Hackers target the user and present a fake website as classification to detect phishing attacks. Different types of
legitimate. The reason is to mislead the user or wrongly features were used for classification like URL-based
present the organization. It is also known as content features, Email features, and Text-based features. An IP
spoofing. The attackers use this strategy to mislead the user address was used as a URL based feature for classification
and to collect information on their server. performed by ML classifies, SVM [20-22], KNN [11], and
BNN. Emails were detected using the Classification
6) Clone Phishing algorithms. The attacker spoofs the email to collect
information. These spoofed emails were detected using a
Biotechnological term 'clone' means producing individuals
the same as the original one. It commonly happens in supervised learning algorithm such as Naïve Bayes [23-25].
Using algorithms, emails were classified, and then spam and
genetic engineering. Clone phishing is another kind of
phishing attack where email is hacked by an opponent party non-spam emails were separated. There were also hyperlink
from the sender or receiver email. The malicious attacker features used for investigation in ML approaches [26-28].
creates alike original email and sends it to the first or second

978-1-7281-5821-1/20/$31.00 ©2020 IEEE 1174

Authorized licensed use limited to: BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE. Downloaded on December 13,2021 at 12:36:58 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Smart Systems and Inventive Technology (ICSSIT 2020)
IEEE Xplore Part Number: CFP20P17-ART; ISBN: 978-1-7281-5821-1

III. M ET HODOLOGY the variables and classify the datasets. The proposed model
This section explains the methodology to detect Phishing was presented in figure 1. To experiment with the phishing
attack using ML, and also explains the proposed framework. website firstly, the dataset was selected. The attributes were
The experiment was carried out by using ML approaches. then analyzed using a feature selection algorithm. The
ML approaches can be applied in two ways. The first one is proposed model has used the REF, Relief-F, IG, and GR
supervised learning, and another is unsupervised learning. algorithm for feature selection. Further, the feature is
Feature selection is crucial for ML algorithms. It reduces the classified between the weakest and most vigorous. Then
redundancy of data which is irrelevant or unnecessary in the PCA was applied for analysis.
data sets. Another statistical method, principal component
analysis (PCA), has been used to identify the components of

Fig. 1 T he proposed framework for phishing detection T able 1. T he phishing dataset

A. Data Acquisition
Data acquisition is essential for data analysis—datasets from
kaggle.com for our research.

B. Data Preprocessing
Data pre-processing is an essential task for the ML
application. It was done from raw data and was formatted
using the data mining technique. A clean and noise-free

978-1-7281-5821-1/20/$31.00 ©2020 IEEE 1175

Authorized licensed use limited to: BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE. Downloaded on December 13,2021 at 12:36:58 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Smart Systems and Inventive Technology (ICSSIT 2020)
IEEE Xplore Part Number: CFP20P17-ART; ISBN: 978-1-7281-5821-1

dataset was needed for analysis of the dataset. Most of the

Used to survey your website and then
dataset contained incomplete and missing values which are
IframeRedirection redirection
filled and completed for ML processing. AgeofDomain Duration of the domain that has existed
Provides important information especially IP
1) Feature Explanation DNSRecording address.
WebsiteT raffic Define web users who visited the website.
Feature selection is an important task for the analysis of the Used to rank the webpage used by google
dataset. Our dataset contained 32 features. Based on the PageRank search engine
information on the features assumption can be made GoogleIndex Add webpages in google search
whether they were distrustful or malicious. The features are LinksPointingToPage Used to rank the webpages
Provides informat ion about all the
explained in table 2. StatsReport downloaded files.
class Contains attributes and behave
T able 2. Feature explanation
Fe ature Name Feature explanation C. Machine Learning Classification Based Modeling
Indexing could be used for the webpage to
Index display in the search engine Supervised learning techniques identify the classification
Using IP Phishing website used IP instead of DNS and used to detect cyber-attacks. In this method, data was
LongURL considered as more than a hundred trained previously to predict the new data. In this research,
LongURL characters two popular machine learning algorithms are used to detect
ShortURL is shorter form of URL. bit.ly is an phishing attacks.
ShortURL example of ShortURL
Symbol@ Special character used in the phishing URL 1) Principal Component Analysis.
Redirecting was used to get intended
Redirecting// destination from the existing link Principal Component Analysis (PCA) was first invented by
Prefix was used the beginning of the original
Karl Pearson in 1901 in the application of mechanics and
word. For example happy become unhappy.
On the other suffix was used the end of the then independently developed by Harold Hotelling in 1930.
PrefixSuffix word. For example, run become running It transforms the large datasets into smaller datasets by
Subdomains were used before the main reducing the dimensionality of the dataset. Increasing data
domain to separate the website into a section, demands a reduction of dimensionality of variables that are
and it was used in many ways and purposes.
For example, blog.mna235.com, the blog was not important to use for prediction. This paper has used
SubDomains a subdomain PCA to reduce the variance by preserving the information
Most of the modern browsers use https for the dataset intact. PCA is much easier to explore the
HT T PS secure website handling. solution and faster to analyze in the machine learning
It was the number of the year that have
algorithm. For that, it has performed to standardize the
registered the domain for a website. It was
paid in advance for registration to start the variable first so that they could turn into the same scale of
Domain Reg Len website. measurement. Secondly, it has performed a covariance
Favicon was an icon used on the website to matrix computation to verify the dependency or correlation
Favicon save space.
of the variables. This feature of space computation was
NonStdPort
It was secure HT T P which was used widely essential to remove the correlated variable, which contained
HT T PSDomainURL with T LS/SSL Protocol. additional information. In the third step, the eigenvectors
It was used to request the resource from the and eigenvalues of the covariance matrix are calculated to
RequestURL server from the client -side. find out principal components. Figure 2 showed the two
It was a clickable text form used to
AnchorURL hyperlink.
principal components about the website, and figure 3
It was used to link at script tag to manipulate showed three dimensions of the correlated data from the
the image, form validation and dynamic dataset.
LinksInScriptT ags content
It was used to process the item in the server
ServerFormHandler from the client.
Info email could be used with its domain or
InfoEmail business website.
AbnormalURL Opposite of normalURL unlikely to occur
It means multiple sources redirect to a single
WebsiteForwarding web address.
It showed information about the system at the
StatusBarCust bottom of the screen.
It prevents saving the content from the
DisableRightClick website. It can be an image or other content.
It appears with a menu on the screen and
UsingPopupWindow disappears after a click

978-1-7281-5821-1/20/$31.00 ©2020 IEEE 1176

Authorized licensed use limited to: BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE. Downloaded on December 13,2021 at 12:36:58 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Smart Systems and Inventive Technology (ICSSIT 2020)
IEEE Xplore Part Number: CFP20P17-ART; ISBN: 978-1-7281-5821-1

5. Use iteration from step 3 and construct a new

decision tree unless you cannot classify any more.
Represent the leaf node as the outcome of the
classification.

Information gain can be obtained using entropy. Split

information and gain ratio were used to select the alternative
attributes which contained numerical values.

3) Random Forest

Random Forest (RF) is one of the robust algorithms in

machine learning. It is a supervised learning algorithm used
for classification and regression. It uses the bagging method
Fig. 2 T wo Principle components
to combine the learning model and average the overall result
for better prediction. RF is used to classify the website
between legitimate and phishing. As the random forest is a
combination of many single trees , it can produce high
ML algorithm Accuracy Recall Precision F1 Score
DT 0.9194 0.9384 0.8804 0.9084
RF 0.9696 0.4216 0.9689 0.5874
accuracy of the result. Firstly it selected samples randomly.
Then a decision tree on each sample of the dataset was built
where the results are obtained from each decision tree. Then
the method was applied to predict the result and to select the
highest voted result for final prediction. RF produced high
Fig. 3. 3D scatter by PCA
accuracy over a single decision tree, even the data was
2) Decision Tree missing. It could overcome the over-fitting problem.

Decision tree (DT) is one of the most popular algorithms in IV. RESULT S AND DISCUSSIONS
machine learning for binary classification. It results from the Phishing attack detection based on feature analysis, data
decision very fast by creating a small tree and can predict analysis on the selected dataset was carried in this paper.
upon training dataset. As its name implies tree, it holds The confusion matrix shows the performance table on
nodes and attribute denotes a test. The branch is the accuracy when compared with the actual classifications in
consequence of the test, and each terminal or end node, the dataset. Accuracy, precision, recall, and F1 score were
which is called leaves are the labels of the classification. used for performance evaluation which was calculated based
Determining the best attribute is the most important in this on the confusion matrix. Confusion matrix used specific
algorithm. Ross Quinlan developed the decision tree ID3 table layout for the projection of the performance, as shown
algorithm. It was primarily used in data mining and in Table 3 and computed according to the following
information theory. Now it is used in machine learning and equations:
natural language processing. The proposed model has used
the ID3 algorithm in this paper to classify the website, Precision =TP/(TP + FP)
whether it was an official or phishing website. The
following steps are followed to get the outcome of the Recall =TP/(TP + FN)
classification of this algorithm:
Accuracy = (TP + TN)/(TP + FP + TN + FN)
1. Start with the training data set. Give it the name 'S'
and it should have attributes and classification F1 Score = 2*((precision*recall)/(precision+recall))

2. Determine the best attribute of the data sets True Positive (TP): Correctly predicted phishing URLs were
detected with the actual phishing URLs. False Negative
3. Divide the 'S' which each have a value of the best (FN): The actual phishing URLs were false classified and
attributes detected as legitimate URLs. False Positive (FP): The actual
legitimate URLs were classified as false values and detected
4. Build a decision tree node which holds the best as phishing URLs. True Negative (TN): The actual class and
attribute. the predicted class was the same as it showed here the actual

978-1-7281-5821-1/20/$31.00 ©2020 IEEE 1177

Authorized licensed use limited to: BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE. Downloaded on December 13,2021 at 12:36:58 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Smart Systems and Inventive Technology (ICSSIT 2020)
IEEE Xplore Part Number: CFP20P17-ART; ISBN: 978-1-7281-5821-1

legitimate URLs were correctly predicted as legitimate REFERENCES

URLs. [1] S. Mahdavifar and A. A. Ghorbani, "DeNNeS: deep embedded
neural network expert system for detecting cyber attacks," (in
Table 3 presented a confusion matrix on the DT algorithm English), Neural Computing & Applications, Article; Early
Access p. 28.
for actual class on phishing and legitimate website with the [2] S. Mishra and D. Soni, "Smishing Detector: A security model to
predicted class, which was classified with phishing and detect smishing through SMS content analysis and URL
legitimate class. In table 4, a confusion matrix on the RF behavior analysis," (in English), Future Generation Computer
Systems-the International Journal of Escience, Article vol. 108,
algorithm was shown, which was classified in the same way pp. 803-815, Jul 2020.
as table 3. In table 5, showed the evaluation matrix on DT [3] K. S. Adewole, T . Hang, W. Q. Wu, H. B. Songs, and A. K.
Sangaiah, "T witter spam account detection based on clustering
and RF for performance analysis. and classification methods," (in English), Journal of
Supercomputing, Article vol. 76, no. 7, pp. 4802-4837, Jul 2020.
T able 3: Confusion Matrix DT Algorithm [4] B. A. Eduardo, F. D. Walter, and S. G. Sandra, "An Experiment
Classified Phishing Classified le gitimate to Create Awareness in People concerning Social Engineering
Actual Phishing 884 58 Attacks," (in Spanish), Ciencia Unemi, Article vol. 13, no. 32,
Actual le gitimate 120 1149 pp. 27-40, Jan-Apr 2020.
[5] S. Farrugia, J. Ellul, and G. Azzopardi, "Detection of illicit
accounts over the Ethereum blockchain," (in English), Expert
T able 4: Confusion Matrix RF Algorithm
Systems with Applications, Article vol. 150, p. 11, Jul 2020, Art.
Classified Phishing Classified legitimate no. 113318.
Actual Phishing 904 38 [6] C. D. Xuan, H. D. Nguyen, and T . V. Nikolaevich, "Malicious
Actual le gitimate 29 1240 URL Detection based on Machine Learning," (in English),
International Journal of Advanced Computer Science and
T able 5: Metrics for Evaluating DT and RF Performance Applications, Article vol. 11, no. 1, pp. 148-153, Jan 2020.
Accuracy is the ratio of absolute predicted class value to the [7] J. C. Gonzalez, V. Garcia-Diaz, E. R. Nunez-Valdez, A. G.
Gomez, and R. G. Crespo, "Replacing email protocols with
all measurement value. blockchain-based smart contracts," (in English), Cluster
Computing-the Journal of Networks Software Tools and
Recall is the ratio of correctly predicted positive Applications, Article; Early Access p. 7.
[8] S. M. Mironova and S. S. Simonova, "Protection of the Rights
measurement to the total measurement of actual yes class and Freedoms of Minors In T he Digital Space," (in Russian),
Russian Journal of Criminology, Article vol. 14, no. 2, pp. 234-
Precision is the ratio of the predicted positive measurement 241, 2020.
to the total predicted positive measurement. [9] D. Sarma, "Security of Hard Disk Encryption," ed, 2012
[10] S. Shabudin, N. S. Sani, K. A. Z. Ariffin, and M. Aliff, "Feature
Selection for Phishing Website Classification," (in English),
F1 score means the weighted average of Precision and International Journal of Advanced Computer Science and
Recall. Hence, this set takes both False Positives and False Applications, Article vol. 11, no. 4, pp. 587-595, Apr 2020.
Negatives into the record to deliver a proportion betwixt [11] A. Zamir et al., "Phishing web site detection using diverse
machine learning algorithms," (in English), Electronic Library,
precision and recall. Article vol. 38, no. 1, pp. 65-80, Jan 2020.
[12] M. A. Adebowale, K. T . Lwin, and M. A. Hossain, "Intelligent
V. CONCLUSION phishing detection scheme using deep learning algorithms," (in
English), Journal of Enterprise Information Management,
A machine learning (ML) based phishing attack detection Article; Early Access p. 20.
was proposed in this paper. The investigation utilizes many [13] G. Sonowal and K. S. Kuppusamy, "PhiDMA - A phishing
detection model with multi-filter approach," (in English),
strategies to identify phishing detection. Standard datasets of Journal of King Saud University-Computer and Information
phishing attacks from kaggle.com were used as input for the Sciences, Article vol. 32, no. 1, pp. 99-112, Jan 2020.
ML algorithms. Two popular machine learning algorithms, [14] N. A. Azeez, B. B. Salaudeen, S. Misra, R. Damasevicius, and
R. Maskeliunas, "Identifying phishing attacks in communication
namely decision tree and random forest, are implemented to networks using URL consistency features," (in English),
analyze and select datasets for classification and detection. International Journal of Electronic Security and Digital
Forensics, Article vol. 12, no. 2, pp. 200-213, 2020.
Principal component analysis (PCA) was applied to identify [15] E. D. Frauenstein and S. Flowerday, "Susceptibility to phishing
and classify the components of the datasets. DT was used to on social network sites: A personality information processing
classify the website and RF was used for classification. model," (in English), Computers & Security, Article vol. 94, p.
18, Jul 2020, Art. no. Unsp 101862.
Finally, the confusion matrix was drawn to evaluate the [16] M. Binjubeir, A. A. Ahmed, M. A. Bin Ismail, A. S. Sadiq, and
performance of these two algorithms . RF had less variance, M. K. Khan, "Comprehensive Survey on Big Data Privacy
Protection," (in English), Ieee Access, Article vol. 8, pp. 20067-
and it could handle the over-fitting problem. The random 20079, 2020.
forest tree achieved an accuracy of 97%. In our future work, [17] G. D. Parra, P. Rad, K. K. R. Choo, and N. Beebe, "Detecting
fishing attacks will be predicted from the logged dataset of Internet of Things attacks using distributed deep learning," (in
English), Journal of Network and Computer Applications,
attacks by using a convolution neural network (CNN). It Article vol. 163, p. 13, Aug 2020, Art. no. 102662.
will be added as a tool for intrusion detection system (IDS). [18] R. T . Pashiri, Y. Rostami, and M. Mahrami, "Spam detection
through feature selection using artificial neural network and
sine-cosine algorithm," (in English), Mathematical Sciences,
Article; Early Access p. 7.

978-1-7281-5821-1/20/$31.00 ©2020 IEEE 1178

Authorized licensed use limited to: BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE. Downloaded on December 13,2021 at 12:36:58 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Smart Systems and Inventive Technology (ICSSIT 2020)
IEEE Xplore Part Number: CFP20P17-ART; ISBN: 978-1-7281-5821-1

[19] S. Anwar et al., "Countering Malicious URLs in Internet of

T hings Using a Knowledge-Based Approach and a Simulated
Expert," (in English), Ieee Internet of Things Journal, Article
vol. 7, no. 5, pp. 4497-4504, May 2020.
[20] A. S. Bozkir and M. Aydos, "LogoSENSE: A companion HOG
based logo detection scheme for phishing web page and Email
brand recognition," (in English), Computers & Security, Article
vol. 95, p. 18, Aug 2020, Art. no. Unsp 101855.
[21] S. E. Raja and R. Ravi, "A performance analysis of Software
Defined Network based prevention on phishing attack in
cyberspace using a deep machine learning with CANT INA
approach (DMLCA)," (in English), Computer Communications,
Article vol. 153, pp. 375-381, Mar 2020.
[22] T . A. T uan, H. V. Long, L. Son, R. Kumar, I. Priyadarshini, and
N. T . K. Son, "Performance evaluation of Botnet DDoS attack
detection using machine learning," (in English), Evolutionary
Intelligence, Article vol. 13, no. 2, pp. 283-294, Jun 2020.
[23] F. Ahmed, et al., "A Combined Belief Rule based Expert
System to Predict Coronary Artery Disease," in 2020
International Conference on Inventive Computation
Technologies (ICICT), 2020, pp. 252-257.
[24] S. Hossain, et al., "A Belief Rule Based Expert System to
Predict Student Performance under Uncertainty," in 2019 22nd
International Conference on Computer and Information
Technology (ICCIT), 2019, pp. 1-6.
[25] S. Hossain et al. "A Critical Comparison between Distributed
Database Approach and Data Warehousing Approach."
International Journal of Scientific & Engineering Research,
Article 5.1 (2014): 196-201.
[26] K. Noor et al., "Performance analysis of a surveillance system to
detect and track vehicles using Haar cascaded classifiers and
optical flow method," 2017 12th IEEE Conference on Industrial
Electronics and Applications (ICIEA), Siem Reap, 2017, pp.
258-263.
[27] S. Hossain, D. Sarma, R. J. Chakma, W. Alam, M. M. Hoque,
and I. H. Sarker, "A Rule-Based Expert System to Assess
Coronary Artery Disease Under Uncertainty," in Computing
Science, Communication and Security, Singapore, 2020, pp.
143-159: Springer Singapore.
[28] S. Hossain, A. Abtahee, I. Kashem, M. M. Hoque, and I. H.
Sarker, "Crime Prediction Using Spatio-T emporal Data," in
Computing Science, Communication and Security, Singapore,
2020, pp. 277-289: Springer Singapore.
[29] H. Alqahtani et al., "Cyber Intrusion Detection Using Machine
Learning Classification T echniques," in Computing Science,
Communication and Security, Singapore, 2020, pp. 121 -131:
Springer Singapore.

978-1-7281-5821-1/20/$31.00 ©2020 IEEE 1179

Authorized licensed use limited to: BIRLA INSTITUTE OF TECHNOLOGY AND SCIENCE. Downloaded on December 13,2021 at 12:36:58 UTC from IEEE Xplore. Restrictions apply.