Web based Application and Collaborative Machine Learning Approaches for Credit Card Fraud Detection

Prusti, Debachudamani (2023) Web based Application and Collaborative Machine Learning Approaches for Credit Card Fraud Detection. PhD thesis.

[img]PDF (Restricted upto 17/04/2026)
Restricted to Repository staff only



With advancement of growing technology, the financial transactions linked to credit card are augmented across online mode. The convenience and popularity of credit card usage leads to the anticipation of fraudulent activities cunningly plotted by the fraudsters. It is a matter of great concern for credit card users as well as financial institutions, to provide credit card facilities for making the transactions free from possible frauds being carried out by fraudsters. To sort out the issue on fraud detection, number of researchers and professionals have applied various strategies including computer algorithms and automation technology. Even though a good number of researchers and analysts have extended their studies in detection of fraudulent transactions by applying various methodologies, still the research track is not so proficient enough due to the privacy concern in research data and variable strategies adopted by fraudsters. Hence, developing a fraud detection system to identify the fraudulent activities is an important area of research to improve the credibility of credit card-based digital transactions. Various model-based approaches have been proposed based on the application of machine learning techniques, deep learning techniques, web service based approach and graph-based algorithms. Critical assessment on performance of various methodologies have been carried out based on the evaluation metrics considered for analysis. Developing a fraud detection model based on machine learning algorithms and ensemble method has been carried out in this study as the first contribution to this thesis. In this study, the application of various classification models has been proposed by implementing machine learning techniques and their performance parameters are critically assessed for detecting fraudulent transactions. Five classification algorithms such as k-nearest neighbor (K-NN), extreme learning machine (ELM), random forest (RF), multilayer perceptron (MLP) and bagging classifier have been implemented because of their improved performance with the considered BankSim and PaySim datasets. We have proposed a predictive classification model by ensemble of five individual machine learning algorithms, as it provides improved predictive performance on different metrics. Second contribution to this thesis highlights on the detection of fraudulent transactions, performed using various deep learning techniques to improve the performance of different measures. For this study, the deep learning techniques such as convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory (LSTM) and autoencoder have been studied on the transactional data BankSim and PaySim to predict the class label as fraudulent or normal. These models are effective enough to identify the high level features vii and the hidden patterns on the data. Each of the four models have been trained with efficient Adam optimizer and the loss function has been used for fitting the binary cross-entropy loss function. Four models are critically assessed to evaluate the parameters such as accuracy, precision, recall, F-measure, AUPR measure and ROC-AUC score for analysis of each model. The third contribution indicates a fraud detection system based on web services by considering two different protocols for the web-based services such as simple object access protocol (SOAP) and representational state transfer (REST). Further, for detecting the fraudulent transactions, these services are associated with five different machine learning techniques such as support vector machine (SVM), multilayer perceptron (MLP), random forest regression, autoencoder and isolation forest due to their improved performance. The performance analysis of each machine learning algorithm associated with SOAP and REST services are critically assessed. The web services have been designed based on concepts of service oriented architecture (SOA) by considering a middleware family of software products i.e., Oracle SOA suite which is very often used by the software architects. Various performance metrics have been evaluated for five machine learning techniques with and without incorporation of the web services. In the fourth contribution, the fraud detection system has been proposed based on the application of graph database model. From the transactional data, a graph model has been developed by applying Neo4j tool and important graph features are extracted from the graph model using different graph algorithms. Subsequently various supervised and unsupervised machine learning algorithms have been applied to detect the fraudulent transactions explicitly with and without the incorporation of graph features in the BankSim and PaySim datasets. Features extracted using different graph algorithms such as degree centrality algorithm (DCA), closeness centrality algorithm (CCA), betweenness centrality algorithm (BCA), PageRank algorithm, label propagation algorithm (LPA) and node clustering coefficient (NCC) are applied with various supervised and unsupervised machine learning techniques individually as well as in combined approach to classify the fraudulent and normal transactions. Graph based methods such as combined graph based approach with machine learning techniques, CGB-GMM and GB-LOF have been proposed to improve the fraud detection by considering the features obtained from the transactional data. With the above contributions, it is intended to develop the fraud detection model with different approaches for identifying the fraud patterns in the transactional data and to study various performance metrics for analysis purpose.

Item Type:Thesis (PhD)
Uncontrolled Keywords:Fraud detection; Machine learning; Deep learning; Web service based approach; Graph-based algorithm; Neo4j tool
Subjects:Engineering and Technology > Computer and Information Science
Engineering and Technology > Computer and Information Science > Information Security
Divisions: Engineering and Technology > Department of Computer Science Engineering
ID Code:10504
Deposited By:IR Staff BPCL
Deposited On:16 Apr 2024 12:08
Last Modified:16 Apr 2024 12:08
Supervisor(s):Rath, Santanu Kumar

Repository Staff Only: item control page