IRJET- Finalize Attributes and using Specific Way to Find Fraudulent Transaction

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 309
Finalize Attributes and using Specific Way to Find Fraudulent
Transaction
Heta Naik1, Prashasti Kanikar2
1Student, Dept of computer Engineering, MPSTME NMIMS, Mumbai, India
2 Professor, Dept. of Computer Engineering, MPSTME NMIMS, Mumbai, India
----------------------------------------------------------------------***---------------------------------------------------------------------
ABSTRACT - Online transactions are increasing because of they make our life easy. Other hand parallel side fraud
transactions are also increasing and that is not good. For finding fraudulent transaction, need some strong mechanism or
algorithm. In this paper first shortlisting algorithms from Literature review. Their 6 shortlisted algorithms. Then finalise one
online transaction dataset. That dataset contains 20 attributes. That all attributes related to transaction directly or indirectly.
For finding attributes priority or effect of attributes use Information gain and gain ratio techniques. After this result finalise 4
category 20, 14, 10 and 7 attributes. This number of attributes category apply on WEKA with shortlisted algorithms. On basis
of this WEKA result shortlisted 4 algorithms for Implementation. AdaBoost, Logistic regression, J48 and Naïve Bayes
algorithms are implemented in python. And final conclusion of research AdaBoost is better for finding fraudulent transactions
compare to other algorithms.
Keywords: Fraud, Credit card, Algorithms, machine learning, attribute selection, Information gain, gain ratio, WEKA,
AdaBoost
INTRODUCTION
Now a day’s number of frauds are increasing. Fraud has many ways to attempt such as Internet Fraud, hacker, Credit card
fraud, Computer crime, etc. Specific credit card fraud occurred via many ways i.e. Skimming, Balance transfer, Account
takeover, online transactions. Etc. Day by day online transactions are increasing because of they makes our life easy. Ticket
booking, glossary shopping, pay online bills. Etc. are done easily online. Other side fraudulent transactions are also
increasing. Find or classify the fraudulent transactions need some mechanism or algorithms. Finding fraudulent
transactions KNN, logistic regression, J48, AdaBoost, outlier, Decision tree, etc. algorithms can be used. There are some
parameters to classify the fraudulent i.e. Time, amount, transaction frequency, etc. Finding fraud need some information
such as Card holder name, address, age, balance, transaction history, expiry date, transaction time, amount, etc. Analyse
fraud or not they need passed analysed results and passed data to analyse and classify the transactions.
LITERATURE REVIEW
Dr. Sanjay Kumar Dwivedi, et.al [14] In this paper they describe Web mining in detail. In web mining there are 3
types of mining: Web content mining, web usage mining and web structure mining. Web content mining is applicable on
Text, Audio, Video, Images and Structured record. Web usage mining apply on Web server logs, Application server logs and
application logs. And 3rd one Web structure mining is apply on Hyperlinks, Document structure and application level. For
the data mining data pre-processing is very important. Data pre-processing have different phases such as data collection,
data cleaning, session identification, user identification and path completion. Web data is inconsistent, noisy and irrelevant
by the nature so that this data can’t be used for analysis or mining. For this data first of all they need to pre-process on
dataset after it can be used for mining and analysis.
Aman Srivastava, et.al [15], In this paper they describe what online transaction is and how they work. They
compare different detection techniques such as Neural networks, Rule Induction, Case-based reasoning, Genetic algorithm,
Inductive logic programming Expert System and Regression. Neural network gave better result. Neural network compare
with human brain and human techniques. Human always try to learn something new and learn from experience. Human
not follow step by step instruction same as neural network. Neural network also lean from past result and training
datasets. This algorithm learn from itself. So, that compare to others Neural network is best to other. This paper suggest
one system or flow to find the fraudulent transactions. This system try to find fraud from merchant side. Payment gateway
have some details such as credit or debit card number, expiry date time, etc. Merchant pass shipping address, amount etc.
then payment gateway send some necessary detail to Fraud detection system. And this Fraud detection system made by
Neural network for train data itself and generate the output. That output is decision of transaction i.e. fraudulent or not.
This decision output send to payment gateway. In the fraud detection system 2 phase such as Learning and testing. Neural
network learn itself from past result and test as Non- fraudulent, Doubtful, Suspicious or Fraudulent transaction. This
technique is very sufficient because merchant know very well about transaction so that number of fraudulent transaction
less compare to others.

S. Benson Edwin Raj, et.al [18], In this paper they describe what is credit card fraud and how to detect credit card
fraud. For detecting credit card fraud give some methods such as Fusion approach using Dempster - Shafer theory and
Bayesian learning, BLAST-SSAHA Hybridization, Hidden Markov Model, Fuzzy Darwinian Detection and Bayesian and
Neural Networks. Dempster – Shafer theory and Bayesian learning is hybrid approach for credit card fraud. This Fusion
approach proposed a fraud detection using Information fusion and Bayesian learning for counter credit card fraud. This
FDS system has four components namely, Rules based filter, Dempster – safer adder, Bayesian learner and Transaction
history dataset. This technique gave high accuracy and high processing speed. They improve detecting rate and reduce
false alarm. BLAST-SSAHA is combination of hybrid BLAST and SSAHA algorithm. BLAST- FDS is a two stage sequence
alignment which are profile analyser (PA) and deviation analyser (DA). PA determine similarity of an incoming sequence
of transaction and DS determine unusual incoming transactions. This algorithm is good. Gave high accuracy and
processing speed also fast enough. Fuzzy Darwinian detection system uses genetic fuzzy logic rules and classify credit card
transaction in ‘Suspicious’ and ‘Non- suspicious’. In this techniques first data cluster in ‘low’, ‘medium’, and ‘high’ range.
This system gave good accuracy and intelligibility level for real data. Conclusion of this paper was Fuzzy Darwinian and
BLAST-SSAHA have very high accuracy in terms of TP and FP. But processing speed is fast enough to enable online
detection of credit card fraud detection in BLAST-SSAH.
Samaneh Sorournejad, .et. al [9], A survey of credit card fraud detection techniques as on data basis and
techniques oriented perspectives Let, in 21st.century credit card is a very important role plays. For using in daily routines,
daily needs, e-shopping and other variants. It should be too many applications for this generation, with using credit card
people has buy whatever needs. With the help of credit card we reduced our external affairs or external corridor time
which is minimized with smartly using of credit card. Credit card is a good for those who really punctual for its personal
economy & finance statistics. At the current situations of the world, banking systems and financial companies are expand
to their facilities to valuable clients for their innovative services without using their money and as on credit period they
given on their client for the same. Therefore credit card is a very harmful and useful part of person’s financial life. From the
bank or financial organizations are given credit card to their client on basis of client’s financial banking record,
government taxation policy’s involvement and other liable priorities which should follow or fulfil their criteria that clients
are eligible for credit card. It means person who is in actual? That also should be defined on basis of credit card holder or
non-holder. Credit card is not get easily to any other person for any uses or any applications. It should be strictly in
banking criteria and its civil report basis they has holder of it. Credit card fraud is also a big issues now a days, some fraud
peoples are using other credit card fraudily and loses are facing some card holders. Its occurs only for a some lack of
knowledge, some mismanagement, and missed their credit card, then fraud people are using benefits of other for personal
needs.
CREDIT CARD FRAUD DATASET
Transactions information are very authenticate. Finding a fraud, need some basic information about credit card and card
holder data.
Attribute Name Data Type Description
Over_draft Qualitative Status of existing checking
account
Credit_usage Numerical Credit usage in month vise
Credit_history Qualitative Credit card history
Purpose Qualitative Purpose of credit card
transaction
Current_balance Numerical Credit Amount
Average_credit_balance Qualitative Average credit amount
Employment Qualitative Present employement
Location Numerical Installemt rate in percentage of
disposable income
Personal_status Qualitative Personal Status and sex
Other_parties Qualitative Other debtors/ guarantors
Recidence_since Numerical Present residence since now
Property_magnitude Qualitative Property
CC_age Numerical CC age in months

Other_payment_plans Qualitative Other installment plans
Housing Qualitative Housing
Existing_credits Numerical Number of exidting credit in
this bank
Job Qualitative Job
Num_dependent Numerical Number of people being liable
to provide maintenance for
Own_telephone Qualitative Telephone is there or not
Foreign_worker Qualitative Foreign workers
This attributes are related to online transactions. Some attributes are directly and some are indirectly connected to
transaction details.
ATTRIBUTE SELECTION
There are many attributes and that attributes contains transaction related data. So, first find the priority wised attributes
from the credit card dataset. Selecting attributes on priority based. Finding attributes priority using two algorithms such
as, Gain Ratio and Information Gain.
Information Gain:
 Information gain = information before splitting – Information after splitting
 It works fine for most cases, unless you have a few variables that have a large number of values (or
classes)
 Information gain is biased towards choosing attributes with a large number of values as root nodes.
Gain Ratio:
 Gain ratio is modification of information gain that reduces its bias and is usually the best option
 This ratio overcome the problem with information gain by taking into account the number of branches
that would result before making the split
Based on both algorithms, finalized attributes from dataset. This two algorithms apply on WEKA. They give
attributes priority rank. Based on attributes priority, make 4 categories such as, 20 attributes, shortlisted first 14
attributes, shortlisted first 10 attributes and shortlisted first 7 attributes.
20 Attributes
(Entire dataset)
Shortlisted
14 Attributes
Shortlisted
10 Attributes
Shortlisted
7 Attributes
1 1 1 1
2 2 2 2
3 3 3 3
4 4 4 4
5 5 5 5
6 6 6 6
7 7 7 20
8 9 12
9 10 13
10 12 20
11 13
12 14
13 15
14 20
15
16
17

18
19
20
SHORTLISTED ALGORITHM USING LITERATURE REVIEW
From literature review there are many algorithms which are used for credit card detection. Such as, K- nearest neighbour,
logistic regression, Outliers. J48, AdaBoost, Naïve Bayes, random tree, etc. From literature survey Naïve Bayes has highest
accuracy 97.92% and Logistic regression has lowest accuracy of 54.86%.
Algorithm Accuracy Advantages Disadvantages
K – nearest
neighbor
97.69% • There is no requirement of
predictive model before
classification.
• Compare to power
methods and other known
anomaly detection
methods, KNN is best
• The accuracy of the
method depends on
the measure of
distance.
• It cannot detect the
fraud at the time of
transaction.
Naïve Bayes 97.92% /
70.13
• High processing and
detection speed/high
accuracy
• Excessive training
need / expensive
Random Tree 94.32% • It gives estimates of what
variables are important in
the classification
• It can handle thousands of
input variables without
variable deletion
• It is over fit for some
datasets with noisy
classification
/regression tasks
Logistic Regression 54.86% • It produces a simple
probability formula for
classification.
• It works well with linear
data for credit card fraud
detection.
• It cannot be applied on
non-linear data
• It is not capable of
handling fraud
detection at the time
of transaction
Outlier • Using less memory and
computation requirements
• Works fast and well on
online large datasets
• It can handle
thousands of input
variables without
variable deletion
AdaBoost 57.73% • It is a powerful classifier
that works well on both
basic and more complex
recognition problems
• It can be sensitive to
noisy data and
outliers.
J48 93.50% • This algorithm use
weighted dataset
• This algorithm can be
payoff but there is
chances to get
different decision
SHORTLISTED ALGORITHMS USING WEKA
Algorithms from Literature survey i.e. Naïve Bayes, J48, Random tree, AdaBoost, Logistic regression and KNN are
implemented on WEKA with Credit card fraud detection.
Selected attributes category wise apply on different algorithms and compare their accuracy.

Algorithms /
Attributes
20 Attributes 14 Attributes 10 Attributes 7 Attributes
Naive Bayes 94.7 95.3 96 96.1
J48 96.3 96.3 96.3 96.3
Random tree 93.7 94.1 93.6 93.6
AdaBoostM1 96.2 96 96.3 96.3
Logistic 95.5 95.1 96.1 96.1
KNN 93.2 94.3 94 93.7
All algorithms run with 20 attributes, 14 attributes, 10 attributes and 7 attributes. Conclude on WEKA result J48 algorithm
has highest accuracy with 20 attributes is 96.3% and lowest accuracy with 20 attributes is KNN 93.2%. From WEKA result
analysed that only J48 algorithm gave constant accuracy with different number of attributes and other algorithms are
increased or decreased with different number of attributes
Based on WEKA result, Naïve Bayes, J48, AdaBoost and Logistic regression algorithms are shortlisted for
implement. Random tree and KNN algorithms has lowest priority with all number of attributes categories.
IMPLEMENTED ALGORITHMS IN PYTHON
Shortlisted WEKA algorithms are Naïve Bayes, J48, AdaBoost and Logistic regression implemented in Python.
Name Naïve Bayes Logistic Regression J48 AdaBoost
Accuracy 83.00% 100.00% 69.93% 100.00%
Time Duration 1.17 3.81 4.62 2.80
Training Testing 70 : 30 70 : 30 70 : 30 70 : 30
Inbuilt Packages Gaussian NB Logistic
Regression
Decision Tree
Classifier
AdaBoost
Classifier
Result of implemented algorithms, AdaBoost and Logistic regression has highest accuracy 100% and J48 has lowest
accuracy 69.96%. AdaBoost and Logistic regression has same accuracy but time duration was different. AdaBoost takes
2.80 seconds and Logistic
Regression takes 3.81seconds. So that, from our Fraud detection flow and Credit card dataset AdaBoost algorithm is more
preferable from others.
ANALYSIS & CONCLUSION
Implemented algorithm with different number of attributes such as 20, 14, 10 and 7 attributes.
This Analysis by Algorithm. After Analysis conclude that AdaBoost algorithm is better than other algorithms.
20 Attributes
10 Attributes
90
95
100
NumberofAttributes
Accuracy
Shortlisted Algorithms
Comparision of shortlisted algorithms for
diffrent attributes using WEKA

s
0
0.5
1
1.5
72
72.5
73
73.5
74
74.5
20 Attributes14 Attributes10 Attributes 7 Attributes
Naive Bayes
Series1 Series2
3.8
4
4.2
4.4
4.6
4.8
5
0
20
40
60
80
J48 Algorithm
Accuracy Time Taken
0
1
2
3
4
5
0
50
100
150
Logistic Regression
Accuracy Time Taken
0
0.5
1
1.5
2
2.5
0
50
100
150
20 Attributes14 Attributes10 Attributes 7 Attributes
AdaBoost
Accuracy Time Taken

REFERENCES
[1] Mukesh Kumar Mishra and Rajashree Dash, “A Comparative Study of Chebyshev Functional Link Artificial Neural
Network, Multi-Layer Perceptron and Decision Tree for Credit Card Fraud Detection”, International Conference on
Information Technology, pp 228 -233, 2014
[2] Pornwatthana Wongchinsri and Werasak Kuratach, “A Survey - Data Mining Frameworks in Credit Card Processing”,
IEEE, 2016
[3] Yufeng Kou, .et. al., “Survey of Fraud Detection Techniques”, International Conference on Networking, Sensing &
Control, pp 749 – 754, 2004
[4] John O. Awoyemi, .et. al., “Credit card fraud detection using Machine Learning Techniques: A Comparative Analysis”
IEEE, 2017
[5] Shiyang Xuan, .et. al., “Random Forest for Credit Card Fraud Detection”, IEEE – 2018
[6] Sahil Dhankhad, .et. al., “Supervised Machine Learning Algorithms for Credit Card Fraudulent Transaction Detection: A
Comparative Study”, IEEE International Conference on Information Reuse and Integration for Data Science, pp 122-125,
2018
[7] R. Brause.et. al., “Neural Data Mining for Credit Card Fraud Detection”
[8] Zahra Kazemi and Houman Zarrabi, “Using deep networks for fraud detection in the credit card transactions”, IEEE 4th
International Conference on Knowledge-Based Engineering and Innovation (KBEI), pp 0630 – 0633, 2017
[9] Samaneh Sorournejad, .et. al., “A Survey of Credit Card Fraud Detection Techniques: Data and Technique Oriented
Perspective”, pp 1 - 26
[10] P.Sreenivas and Dr.C.V.Srikrishna, “An Analytical approach for Data Pre-processing”
[11] Mine Isik, M.hamdi, et.al, “Improving a credit card fraud detection system using genetic algorithm”, International
Conference on Networking and Information Technology, pp 437 – 440, 2010
[12] Kuldeep Randhawa, Chu Kiong Loo, et.al, “Credit card fraud detection using AdaBoost and majority voting, IEEE
Access, pp- 1 – 8, vol XX, 2017
[13] Dr. Sanjay Kumar Dwivedi and Bhupesh Rawat, “A Review Paper on Data pre-processing: A Critical Phase in Web
Usage Mining Process”, International Conference on Green Computing and Internet of Things (ICGCIoT), pp 506-510, 2015
[14] Aman Srivastava, Mugdha Yadav, et.al, “Credit Card Fraud Detection at Merchant Side using Neural Networks”,
International Conference on Computing for Sustainable Global Development (INDIACom), pp- 667- 670, 2016 IEEE
[15] Chun-Hua JU and Na Wang, “Research on Credit Card Fraud Detection Model Based on Similar Coefficient Sum”, First
International Workshop on Database Technology and Applications, IEEE computer Society, pp 295 – 298, 2009
[16] Krishna Modi and Reshma Dayma, “Review On Fraud Detection Methods in Credit Card Transactions”, International
Conference on Intelligent Computing and Control (I2C2'17), 2017
[17] S. Benson Edwin Raj and A. Annie Portia, “Analysis on Credit Card Fraud Detection Methods”, International
Conference on Computer, Communication and Electrical Technology – ICCCET, pp 152 – 156, March - 2011
[18] http://weka.8497.n7.nabble.com/file/n23121/credit_fruad.arff

IRJET- Finalize Attributes and using Specific Way to Find Fraudulent Transaction

More Related Content

What's hot (20)

Similar to IRJET- Finalize Attributes and using Specific Way to Find Fraudulent Transaction (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Finalize Attributes and using Specific Way to Find Fraudulent Transaction