Enhanced Fraudulent Detection Using Isolation Forest and Multi-Cluster Deep Learning

The anomaly detection problem has received increasing research interest due to the negative effects of fraud on several essential systems. Since Iraq is currently moving towards activating electronic financial transactions in all government ministries and private trade, this follows an increase in the risks of financial fraud. This research aims to improve the ability to identify fraudulent financial operations based on a multistage classification model that utilizes several machine learning techniques. It focused on avoiding the outlier instances that can affect the performance of the learning process by utilizing Isolation Forest. The implementation of the proposed model indicates that the ensemble size has no significant impact on its performance while increasing the number of clusters has led to a decline in performance. The experimental results with real datasets produced an F1-score of 99.097 compared to 80.5 and 74.65 with typical DNN, K-NN, and confirmed its preference compared to many popular classifiers and recent research articles.