Fraud Detection and Prevention in Banking Financial Transaction with machine learning using R

Abstract

Fraud is an intentional act of deception involving financial transaction for the purpose of personal gain. With the increased number of online transaction, frauds have also increased. In banking sector detecting fraud is important to keep customers’ money safe and also to reduce the losses from fraud and keep company profitable. Traditional fraud detection methods are no more sufficient in detecting frauds so banks are adopting machine learning based models. One major problem with the financial transaction data is its skewness. Performance of any model depends on dataset and the technique applied. This paper has compared seven machine learning models (logistics regression, random forest, XGBoost, DBscan, Artificial neural network, isolation forest, Principle component analysis with Support vector machine) with the help of several parameters as accuracy, sensitivity, specificity, precision, balanced accuracy (BCR), Matthews correlation coefficient (MCC), kappa value. The study was done for a period of four months on Paysim synthetic dataset of mobile money transactions published on kaggle. The machine learning models were created using R and data analysis was done with the help of tableau. Post analysis It is found that random forest and XGBoost is providing better result than other models.