Special Issue - 2019 International Journal of Engineering Research & Technology (IJERT)

ISSN: 2278-0181
RTICCT - 2019 Conference Proceedings

Credit Card Fraud Detection using Adaboost

and Majority Voting
Mr. S. Siva Prakash M.E.,(Ph.D.),1 Ahubhakumar2,S.Appash3, J.Cibiragul4
1 2,3,4
Professor, UG Students,
Department of Computer Science and Engineering. Department of Computer Science and Engineering.
K.S.R. College of Engineering, K.S.R. College of Engineering,
Tiruchengode, India. Tiruchengode, India.

Abstract: - Credit card fraud is a serious problem in financial effective fraud detection system to reduce or eliminate fraud cases
services. Billions of dollars are lost due to credit card fraud is important. There have been various studies on credit card fraud
every year. There is a lack of research studies on analyzing detection. Machine learning and related methods are most
real-world credit card data owing to confidentiality issues. In commonly used, which include artificial neural networks, rule-
this paper, machine learning algorithms are used to detect induction techniques, decision trees, logistic regression, and
credit card fraud. Standard models are firstly used. Then, support vector machines [1]. These methods are used either
hybrid methods which use AdaBoost and majority voting standalone or by combining several methods together to form
methods are applied. To evaluate the model efficacy, a hybrid models.
publicly available credit card data set is used. Then, a real-
world credit card data set from a financial institution is In this paper, a total of twelve machine learning
analyzed. In addition, noise is added to the data samples to algorithms are used for detecting credit card fraud. The
further assess the robustness of the algorithms. The algorithms range from standard neural networks to deep learning
experimental results positively indicate that the majority models. They are evaluated using both benchmark and realworld
voting method achieves good accuracy rates in detecting fraud credit card data sets. In addition, the AdaBoost and majority
cases in credit cards. voting methods are applied for forming hybrid
models. To further evaluate the robustness and reliability of the
1. INTRODUCTION models, noise is added to the real-world data set. The key
contribution of this paper is the evaluation of a variety of machine
Fraud is a wrongful or criminal deception aimed to learning models with a real-world credit card data set for fraud
bring financial or personal gain. In avoiding loss from fraud, two detection. While other researchers have used various methods on
mechanisms can be used: fraud prevention and fraud detection. publicly available data sets, the data set used in this paper are
Fraud prevention is a proactive method, where it stops fraud from extracted from actual credit card transaction information over
happening in the first place. On the other hand, fraud detection is three months.
needed when a fraudulent transaction is attempted by a fraudster.
The organization of this paper is as follows. In Section
Credit card fraud is concerned with the illegal use of II, related studies on single and hybrid machine learning
credit card information for purchases. Credit card transactions can algorithms for financial applications is given. The machine
be accomplished either physically or digitally. In physical learning algorithms used in this study are presented in Section III.
transactions, the credit card is involved during the transactions. In The experiments with both benchmark and realworld credit card
digital transactions, this can happen over the telephone or the data sets are presented in Section IV. Concluding remarks and
internet. Cardholders typically provide the card number, expiry recommendations for further work
date, and card verification number are given in Section V.
through telephone or website.
2. RELATED WORK
With the rise of e-commerce in the past decade, the use
of credit cards has increased dramatically. The number of credit In this section, single and hybrid machine learning
card transactions in 2011 in Malaysia were at about 320 million, algorithms for financial applications are reviewed. Various
and increased in 2015 to about 360 million. Along with the rise of financial applications from credit card fraud to financial statement
credit card usage, the number of fraud cases have been constantly fraud are reviewed.
increased. While numerous authorization techniques have been in
place, credit card fraud cases have not hindered effectively. A. SINGLE MODELS
Fraudsters favour the internet as their identity and location are For credit card fraud detection, Random Forest (RF),
hidden. The rise in credit card fraud has a big impact on the Support Vector Machine, (SVM) and Logistic Regression (LOR)
financial industry. The global credit card fraud in 2015 reached to were examined in. The data set consisted of one-year transactions.
a staggering USD $21.84 billion. Data under-sampling was used to examine the algorithm
performances, with RF demonstrating a better performance as
Loss from credit card fraud affects the merchants, compared with SVM and LOR [6]. An Artificial Immune
where they bear all costs, including card issuer fees, charges, and Recognition System (AIRS) for credit card fraud detection was
administrative charges. Since the merchants need to bear the loss, proposed in. AIRS is an improvement over the standard AIS
some goods are priced higher, or discounts and incentives are model, where negative selection was used to achieve higher
reduced. Therefore, it is imperative to reduce the loss, and an

Volume 7, Issue 01 Published by, www.ijert.org 1

Special Issue - 2019 International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
RTICCT - 2019 Conference Proceedings

precision. This resulted in an increase of accuracy by 25% and Testing credit card FDSs using real data set is a difficult
reduced system response time by 40%. task.
The fraud has to be deducted in real time and the number of
A credit card fraud detection system was proposed in, false alert.
which consisted of a rule-based filter, Dumpster–Shafer adder, 3.2 PROPOSED SYSTEM
transaction history database, and Bayesian learner. The Total of twelve machine learning algorithms are used for
Dempster–Shafer theory combined various evidential information detecting credit card fraud.
and created an initial belief, which was used to classify a The algorithms range from standard neural networks to deep
transaction as normal, suspicious, or abnormal. If a transaction learning models.
was suspicious, the belief was further evaluated using transaction In addition, the AdaBoost and majority voting methods are
history from Bayesian learning. applied for forming hybrid models.
The key contribution of this paper is the evaluation of a
Simulation results indicated a 98% true positive rate. A variety of machine learning models with a real-world credit card
modified Fisher Discriminant function was used for credit card data set for fraud detection.
fraud detection in. The modification made the traditional
functions to become more sensitive to important instances. A SYSTEM ARCHITECTURE DIAGRAM –
weighted average was utilized to calculate variances, which BASEPAPER
allowed learning of profitable transactions. The results from the
modified function confirm it can eventuate more profit.

B. HYBRID MODELS
Hybrid models are combination of multiple individual
models. A hybrid model consisting of the Multilayer Perceptron
(MLP) neural network, SVM, LOR, and Harmony Search (HS)
optimization was used in to detect corporate tax evasion. HS was
useful for finding the best parameters for the classification
models. Using data from the food and textile sectors in Iran, the
MLP with HS optimization acquired the highest accuracy rates at
90.07%. A hybrid clustering system with outlier detection
capability was used to detect fraud in lottery and online games.
The system aggregated online algorithms with statistical
information from the input data to identify a number of fraud
types. The training data set was compressed into the main
memory while new data samples could be incrementally added
into the stored datacubes. The system achieved a high detection
rate at 98%, with a 0.1% false alarm rate.

To tackle financial distress, clustering and classifier 3.2.1 Advantages Of Proposed System
ensemble methods were used to form hybrid models in. The SOM The system is very fast due to AdaBoost Technique.
and k-means algorithms were used for clustering, while LOR, Effective Majority Voting techniques.
MLP, and DT were used for classification. Based on these
methods, a total of 21 hybrid models with different combinations 4.0 METHODOLOGY
were created and evaluated with the data set. The SOM with the
MLP classifier performed the best, yielding the highest prediction MACHINE LEARNING ALGORITHMS
accuracy. An integration of multiple models, i.e. RF, DR, Roush A total of twelve algorithms are used in this
Set Theory (RST), and back-propagation neural network was used experimental study. They are used in conjunction with the
in to build a fraud detection model for corporate financial AdaBoos
statements. Company financial statements in period of 1998 to
2008 were used as the data set. The results showed that the hybrid ALGORITHMS
model of RF and RST gave the highest classification accuracy. Naïve Bayes (NB) uses the Bayes’ theorem with strong
or naïve independence assumptions for classification. Certain
3. PROBLEM STATEMENT features of a class are assumed to be not correlated to others. It
requires only a small training data set for estimating the means
3.1 EXISTING MODEL and variances is needed for classification.
Three methods to detect fraud are presented.
Firstly, clustering model is used to classify the legal and The presentation of data in form of a tree structure is
fraudulent transaction using data parameter value. useful
Secondly, Gaussian mixture model past behavior and current for ease of interpretation by users. The Decision Tree (DT) is a
behavior can be calculated to detect any abnormalities from the collection of nodes that creates decision on features connected to
past behavior. certain classes. Every node represents a splitting rule for a feature.
Lastly, Bayesian networks are used to describe the statistics New nodes are established until the stopping criterion is met. The
of a particular user and the statistics of different fraud scenarios. class label is determined based on the majority of samples that
belong to a particular leaf. The Random Tree (RT) operates as a
3.1.1 Drawbacks
DT operator, with the exception that in each split, only a random
The high amount of losses due to fraud and the
subset of features is available. It learns from both nominal and
awareness of the relation between loss and the available limit has
numerical data samples. The subset size is defined using a subset
to be reduced.
ratio parameter.

Volume 7, Issue 01 Published by, www.ijert.org 2

Special Issue - 2019 International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
RTICCT - 2019 Conference Proceedings

computational underlying domain

The Random Forest (RF) creates an ensemble of power; suitable for information;
random trees. The user sets the number of trees. The resulting realtime retraining
model employs voting of all created trees to determine the final operations. is required for new
types of fraud
classification outcome. The Gradient Boosted Tree (GBT) is an cases.
ensemble of classification or regression models. It uses forward- Neural Network Suitable for binary Need a high
learning ensemble models, which obtain predictive results using classification computational
gradually improved estimations. Boosting helps improve the tree problems, and power, unsuitable
accuracy. The Decision Stump (DS) generates a decision tree with widely for
a single split only. It can be used in classifying uneven data sets. used for fraud real-time
detection. operations; re-
The MLP network consists of at least three layers of training is required
for new
nodes, i.e., input, hidden, and output. Each node uses a non-linear
types of fraud
activation function, with the exception of the input nodes. It uses cases.
the supervised backpropagation algorithm for training. The Linear Provide optimal Sensitive to outliers
version of MLP used in this study is able to adjust the learning Regression results when the and limited to
rate and hidden layer size automatically during training. It uses an relationship numeric values
ensemble of networks trained in parallel with different rates and between only.
number of hidden units. independent and
dependent variables
The Feed-Forward Neural Network (NN) uses the are almost linear.
Linear Provide optimal Sensitive to outliers
backpropagation algorithm for training as well. The connections
Regression results when the and limited to
between the units do not form a directed cycle, relationship numeric values
and information only moves forward from the input nodes to the between only.
output nodes, through the hidden nodes. Deep Learning (DL) is independent and
based on an MLP network trained using a stochastic gradient dependent variables
descent with backpropagation. It contains a large number of are almost linear.
hidden layers consisting of neurons with tanh, rectifier, and Logistic Easy to implement, Poor classification
maxout activation functions. Every node captures a copy of the Regression and historically performances as
used for fraud compared with
global model parameters on local data, and contributes
detection. other data mining
periodically toward the global model using model averaging. methods.
Support Vector Able to solve non- Not easy to process
Linear Regression (LIR) models the relationship Machine linear classification the results due to
between scalar variables by fitting a linear equation to the problems; require transformation of
observed data. The relationships are modelled using linear a low the input data.
predictor functions, with unknown model parameters estimated computational
from the data set. The Akaike criterion, a measure of relative power; suitable for
goodness of fit for a statistical model, is used for model selection. real-time
operations.
Logistic Regression (LOR) can handle data with both nominaland
numerical features. It estimates the probability of a binary
response based on one or more predictor features. 6. CONCLUSION

The SVM can tackle both classification and regression A study on credit card fraud detection using machine
data. SVM builds a model by assigning new samples to one learning algorithms has been presented in this paper. A number of
category or another, creating a non-probabilistic binary linear standard models which include NB, SVM, and DL have been
classifier. It represents the data samples as points in the space used in the empirical evaluation. A publicly available credit card
mapped so such that the data samples of different categories can data set has been used for evaluation using individual (standard)
be separated by a margin as wide as possible. A summary of the models and hybrid models using AdaBoost and majority voting
strengths and limitations of the methods discussed earlier is given combination methods. The MCC metric has been adopted as a
in Table I. performance measure, as it takes into account the true and false
TABLE I positive and negative predicted outcomes. The best MCC score is
STRENGTHS AND LIMITATIONS OF MACHINE 0.823, achieved using majority voting. A real credit card data set
LEARNING METHODS from a financial institution has also been used for evaluation. The
same individual and hybrid models have been employed. A
Model Strengths Strengths Limitations perfect MCC score of 1 has been achieved using AdaBoost and
Limitations majority voting methods. To further evaluate the hybrid models,
Bayesian Good for binary Need good noise from 10% to 30% has been added into the data samples. The
classification understanding of majority voting method has yielded the best MCC score of 0.942
problems; efficient typical and for 30% noise added to the data set. This shows that the majority
use of abnormal voting method is stable in performance in the presence of noise.
computational behaviors for
For future work, the methods studied in this paper will
resources; suitable different types of
for real-time fraud cases be extended to online learning models. In addition, other online
operations. learning models will be investigated. The use of online learning
Trees Easy to understand Potential of over- will enable rapid detection of fraud cases, potentially in real-time.
and implement; the fitting if the This in turn will help detect and prevent fraudulent transactions
procedures training set does not before they take place, which will reduce the number of losses
require a low represent the incurred every day in the financial sector.

Volume 7, Issue 01 Published by, www.ijert.org 3

Special Issue - 2019 International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
RTICCT - 2019 Conference Proceedings

7. REFERENCE Soft Computing, to be published. DOI:

10.1016/j.asoc.2017.07.027.
[22] S. Subudhi and S. Panigrahi, “Use of optimized Fuzzy C-Means
[1] Y. Sahin, S. Bulkan, and E. Duman, “A cost-sensitive decision
clustering and supervised classifiers for automobile insurance
tree approach for fraud detection,” Expert Systems with
fraud detection,” Journal of King Saud University-Computer and
Applications, vol. 40, no. 15, pp. 5916–5923, 2013.
Information Sciences, to be published. DOI:
[2] A. O. Adewumi and A. A. Akinyelu, “A survey of machine-
10.1016/j.jksuci.2017.09.010.
learning and nature-inspired based credit card fraud detection
[23] M. Seera, C. P. Lim, K. S. Tan, and W. S. Liew, “Classification
techniques,” International Journal of System Assurance
of transcranial Doppler signals using individual and ensemble
Engineering and Management, vol. 8, pp. 937–953, 2017.
recurrent neural networks,” Neurocomputing, vol. 249, pp. 337-
[3] A. Srivastava, A. Kundu, S. Sural, A. Majumdar, “Credit card
344, 2017.
fraud detection using hidden Markov model,” IEEE Transactions
[24] E. Duman, A. Buyukkaya, and I. Elikucuk, “A novel and
on Dependable and Secure Computing, vol. 5, no. 1, pp. 37–48,
successful credit card fraud detection system Implemented in a
2008.
Turkish Bank,” In IEEE 13th International Conference on Data
[4] The Nilson Report (October 2016) [Online]. Available:
Mining Workshops (ICDMW), pp. 162–171, 2013.
https://www.nilsonreport.com/upload/content_promo/The_Nilson
[25] C. Phua, K. Smith-Miles, V. Lee, and R. Gayler, “Resilient
_Report_10-17-2016.pdf
identity crime detection,” IEEE Transactions on Knowledge and
[5] J. T. Quah, and M. Sriganesh, “Real-time credit card fraud
Data Engineering, vol. 24, no. 3, pp. 533–546, 2012.
detection using computational intelligence,” Expert Systems with
Applications, vol. 35, no. 4, pp. 1721–1732, 2008.
[6] S. Bhattacharyya, S. Jha, K. Tharakunnel, and J. C., “Data mining
for credit card fraud: A comparative study,” Decision Support
Systems, vol. 50, no. 3, pp. 602–613, 2011.
[7] N. S. Halvaiee and M. K. Akbari, “A novel model for credit card
fraud detection using Artificial Immune Systems,” Applied Soft
Computing, vol. 24, pp. 40–49, 2014.
[8] S. Panigrahi, A. Kundu, S. Sural, and A. K. Majumdar, “Credit
card fraud detection: A fusion approach using Dempster–Shafer
theory and Bayesian learning,” Information Fusion, vol. 10, no. 4,
pp. 354–363, 2009.
[9] N. Mahmoudi and E. Duman, “Detecting credit card fraud by
modified Fisher discriminant analysis,” Expert Systems with
Applications, vol. 42, no. 5, pp. 2510–2516, 2015.
[10] D. Sánchez, M. A. Vila, L. Cerda, and J. M. Serrano,
“Association rules applied to credit card fraud detection,” Expert
Systems with Applications, vol. 36, no. 2, pp. 3630–3640, 2009.
[11] E. Duman and M. H. Ozcelik, “Detecting credit card fraud by
genetic algorithm and scatter search,” Expert Systems with
Applications, vol. 38, no. 10, pp. 13057–13063, 2011.
[12] P. Ravisankar, V. Ravi, G. R. Rao, and I. Bose, “Detection of
financial statement fraud and feature selection using data mining
techniques,” Decision Support Systems, vol. 50, no. 2, pp. 491–
500, 2011.
[13] E. Kirkos, C. Spathis, and Y. Manolopoulos, “Data mining
techniques for the detection of fraudulent financial statements,”
Expert Systems with Applications, vol. 32, no. 4, pp. 995–1003,
2007.
[14] F. H. Glancy and S. B. Yadav, “A computational model for
financial reporting fraud detection,” Decision Support Systems,
vol. 50, no. 3, pp. 595–601, 2011.
[15] D. Olszewski, “Fraud detection using self-organizing map
visualizing the user profiles,” Knowledge-Based Systems, vol. 70,
pp. 324–334, 2014.
[16] J. T. Quah and M. Sriganesh, “Real-time credit card fraud
detection using computational intelligence,” Expert Systems with
Applications, vol. 35, no. 4, pp. 1721–1732, 2008.
[17] E. Rahimikia, S. Mohammadi, T. Rahmani, and M. Ghazanfari,
“Detecting corporate tax evasion using a hybrid intelligent
system: A case study of Iran,” International Journal of
Accounting Information Systems, vol. 25, pp. 1–17, 2017.
[18] I. T. Christou, M. Bakopoulos, T. Dimitriou, E. Amolochitis, S.
Tsekeridou, and C. Dimitriadis, “Detecting fraud in online games
of chance and lotteries,” Expert Systems with Applications, vol.
38, no. 10, pp. 13158–13169, 2011.
[19] C. F. Tsai, “Combining cluster analysis with classifier ensembles
to predict financial distress” Information Fusion, vol. 16, pp. 46–
58, 2014.
[20] F. H. Chen, D. J. Chi, and J. Y. Zhu, “Application of Random
Forest, Rough Set Theory, Decision Tree and Neural Network to
Detect Financial Statement Fraud–Taking Corporate Governance
into Consideration,” In International Conference on Intelligent
Computing, pp. 221–234, Springer, 2014.
[21] Y. Li, C. Yan, W. Liu, and M. Li, “A principle component
analysisbased random forest with the potential nearest neighbor
method for automobile insurance fraud identification,” Applied

Volume 7, Issue 01 Published by, www.ijert.org 4