Fraud Detection and Challenges in Digital Transactions using Bayesian Optimized Random Forest Classifier

Abstract

With technological and economic advances, which facilitated the communication process and increased purchasing power, credit card transactions have become the main payment method in national and international retail. In this regard, the increase in the number of credit card transactions is crucial for generating more opportunities for fraudsters to produce new forms of fraud, which results in great losses for the financial system. This fact arouses the interest of fraudsters. The card market sees fraud as operating costs, which are passed on to consumers and society in general. Still, the high volume of transactions and the need to fight fraud open space for the application of artificial intelligence and machine learning techniques. This research work highlights the importance and challenges faced while using auto monitoring to detect credit card fraud to prevent different risks to our assets. Automated learning techniques have proven to be the solution to supervised learning. This work identifies techniques such as random forests and Bayesian optimized random forest classifiers as the best techniques according to related works. Binary particle swarm optimization algorithm is used for selection of features from the open dataset provided by Kaggle. Sensitivity, precision, f-score and accuracy are used as a performance evaluation tool to find the best parameterization of both the techniques.