Adaptive financial fraud detection system using modified rain optimization algorithm

Abstract

This paper describes an adaptive financial fraud detection system that uses a modified rain optimization algorithm. There are three phases to the proposed approach. The first phase is data pre-processing using Synthetic Minority Oversampling Technique (SMOTE). The second phase includes feature selection on pre-processed data using the Modified Rain Optimization Algorithm (MROA). To evaluate features based on cost functions, MROA uses a stochastic diffusion method, which allows for iterative feature merger or absorption to maximize selection. The feature selection chooses relevant features which will improve the proposed model's performance. In the third phase, the Deep Convolutional Capsule Auto Encoder is used to classify patterns from the selected features. With the advancement of deep learning and the detection capability of Capsule Auto Encoder, the model can easily detect the fraud from the data input. The performance of the proposed system is evaluated with a synthetic financial dataset available on Kaggle and compared to existing methods. In results section, the proposed model is compared with various existing models. The proposed model attained an accuracy of 97.91%, precision of 97.64%, recall of 97.93%, F1-score of 97.81%, specificity of 97.63% and geometric mean of 97.78%, respectively. The results obtained empirically demonstrate that the proposed system performs superiorly with practical implications for financial decision-making.