Hybridizing graph‐based Gaussian mixture model with machine learning for classification of fraudulent transactions

• Published in: Computational intelligence Vol. 38; no. 6; pp. 2134 - 2160
• Main Authors: Prusti, Debachudamani, Behera, Ranjan Kumar, Rath, Santanu Kumar
• Format: Journal Article
• Language: English
• Published: Hoboken Blackwell Publishing Ltd 01-12-2022
• Subjects: Algorithms; Artificial neural networks; classification algorithm; Credit card fraud; Credit cards; Data analysis; Feature extraction; Fraud; Gaussian mixture model; graph feature; Graph theory; Machine learning; Neo4j$$ Neo4j $$ database; Nodes; Probabilistic models; Support vector machines
• ISSN: 0824-7935; 1467-8640
• DOI: 10.1111/coin.12561

Summary
It has been observed that a good number of financial organizations often face a number of threats due to credit card fraud that affects consistently to the card holder as well as the organizations. This is one of the fastest‐growing frauds of its kind and the most emerging problems for the institutions to prevent. A number of researchers and analysts have shown interest to work on this area in order to identify such issues in an effective manner by applying various supervised as well as unsupervised learning approaches. In this assessment, three classification techniques such as support vector machine (SVM), k‐nearest neighbor (k‐NN), and extreme learning machine (ELM) that come under supervised learning category are applied to the BankSim data to categorize the normal and fraudulent class transactions in credit card. These algorithms are incorporated with the graph features extracted from the dataset by using a database tool Neo4j. The nodes of the graph represent the transactional data samples and the edges create relationships among the nodes to find the patterns of data using connected data analysis. k‐fold cross validation approach in Gaussian mixture model (GMM) has been applied for classification of the credit card transaction data in a single distribution. Further, a combined graph‐based Gaussian mixture model (CGB‐GMM) has been proposed to effectively detect the fraudulent instances in credit card transactions with the application of graph algorithms such as degree centrality, LPA, page rank, and so forth. Each of the learning algorithms are implemented with and without the application of graph algorithms and their performances are assessed empirically for analysis.