Fraud Detection Using Apache Spark

Fraud detection methods are continuously developed to defend criminals. They allow us to identify quickly and easily the frauds. In this work, we will focus on the problem of fraud detection in banking transactions. A single algorithm may not be suitable for every problem. Therefore, selecting an algorithm that performs best in a given situation is very crucial. In this work, we give a comparative analysis of four algorithms: Simple Anomaly detection algorithm, Decision Tree algorithm, Random Forest algorithm and Naive Bayes algorithm. We use the machine learning library (MLlib) of Apache Spark to handle credit card fraud detection. The data used in our simulation is generated randomly following a normal distribution, this data includes two features Price and Distance that allow us to distinguish anomalies and valid transactions. The performance is analysed based on the parameters of the Total Running Time and the Accuracy. The results proved that the Random Forest algorithm gave the best results and the simple anomaly detection algorithm gave the worst results.