Imbalance Data Classification to Identify Fraudulent Transactions

Abstract

Because of the expansion of social media and globalization now a days, peta byte scale of data is being generated in every second. Data mining is the process of extracting knowledge from this huge amount of data. Data mining applications are becoming more useful and key pre-requisite for any kind of business scenarios. However, for certain applications is supervised learning, lack of sufficient data for certain classes creates data imbalance problem. For example, in a credit card fraud detection application, most of the transactions are not fraud and few of them are fraud. In our research, we have applied some classification techniques on an imbalanced data set. We have tested synthetic data from a financial payment system because it is a great challenge to obtain real dataset. Synthetic data is artificially constructed which mimics real world events. We have tested Decision tree, Support Vector Machine, Artificial Neural Network and Adaboost algorithms to treat with class imbalance problem. Among these algorithms, we find promising accuracy from Adaboost compared of others. So in this paper, our main target is that for an imbalance dataset which classification algorithm performs better.