Comparative Analysis of Data Mining Techniques for Fraud Detection (A Case Study of Branchless Banking)

Abstract

Data mining algorithms have been using since few years in financial institutions like banks, insurance organizations, etc, and these organizations are using applications of data mining techniques in prediction of business collapse, marketing analysis and fraud detection. In this study our objective is to provide a comparative analysis and find the most suitable techniques of data mining for fraud detection in the area of branchless banking on certain comparison criteria. We have used few different mining algorithms like decision tree, association rules, clustering, naïve bayes and neural network. Our other objective is to find out the comparison criteria, through which we compare these algorithms and that criteria are training volume (small dataset) against quality patterns level, model creation Time, ease of implementation, ease of presentation, extensibility, efficiency, simplicity, training volume (large dataset) against quality patterns level, popularity. In the end we have suggested the most suitable algorithms for fraud detection on branches bank.