Detection of Fraudulent Transactions from Highly Imbalanced Dataset using Different ML Classifiers

Abstract

Nowadays, there is faster rate of increase in the use of card payment and online payments. There is most common issue with such card or online transaction is fraud possibility and there is need to understand cybersecurity concern. Globally it is observed that 43-billion-dollar loss got in 5 years. So, financial organizations such as banks need to identify such fraud transaction details and performance of machine learning algorithms. In this application there is performance analysis of different machine learning classifiers including unsupervised and supervised machine learning classifiers. The dataset used has highly imbalanced data as supervised transactions are very less compared to normal/no-fraud transactions. Unsupervised machine learning algorithms performs superior as compare to supervised machine learning classifiers for credit card fraud detection from highly imbalanced dataset. So, this machine learning based fraud transaction helps in preventing/stopping abnormal transactions and allowing normal transactions. Performance can be measured using different parameters such as balanced accuracy, precision, recall, negative prediction rate, etc. It is observed with result analysis that unsupervised machine learning classifiers such as K-means algorithm, isolation forest algorithm, local outlier algorithm performs much superior than supervised machine learning algorithms such as Support vector machine algorithm, random forest algorithm, naïve bayes algorithms, etc.