Analysis of Banking Data Using Machine Learning with Hadoop for Customer Retention and Fraud Detection

Abstract

Now days, banking industries are facing various challenges such as customer retention, fraud detection, risk management and customer segmentation. On other side banking industry generates massive volume of data every day. It can be possible to provide solutions to these problems faced by the banks using available different information. The data analysis and Machine Learning (ML) help to extract hidden knowledge from the raw data which useful to take real world decisions. In this work, we proposed system which provides solution to problems of customer retention and fraud detection. We used supervised learning techniques namely Artificial Neural Network (ANN), Support Vector Machine (SVM), Deep Neural Network (DNN) and ID3 to analyze bank customer data and German credit fraud data to perform customer retention and fraud detection respectively and performance comparison of these algorithms is carried out. The decision tree algorithm, ID3 is implemented in MapReduce environment to process large volume of data with less execution time. It provides cost efficient and time effective solution. The experimental results show that ANN, SVM, DNN and ID3 give accuracy of 98%, 92%, 97% and 96% respectively, when applied on bank customer data and 72%, 72%, 76% and 67% accuracy respectively for German credit fraud data. This indicates that, the ANN, ID3 and DNN give better performance than SVM to perform customer retention and DNN works better to perform fraud detection. It is also noticed that selected attributes in the datasets are able to perform classification successfully. The proposed system provides an efficient solution for customer retention by classifying active and inactive customers accurately. Also system able to give solution for fraud detection by identifying fraudulent and not fraudulent credits, so it can helps to predict risk associated with the credits. The proposed system helps to improve business profit.