CNN-Bidirectional LSTM based Approach for Financial Fraud Detection and Prevention System

Detecting fraudulent activity has become a pressing issue in the ever-expanding realm of financial services, which is vital to ensuring a positive ecosystem for everyone involved. Traditional approaches to fraud detection typically rely on rule-based algorithms or manually pick a subset of attributes to perform prediction. Yet, users have complex interactions and always display a wealth of information when using financial services. These data provide a sizable Multiview network that is underutilized by standard approaches. The proposed method solves this problem by first cleaning and normalizing the data, then using Kernel principal component analysis to extract features, and finally using these features to train a model with CNN-BiLS TM, a neural network architecture that combines the best parts of the Bidirectional Long Short-Term Memory (BiLS TM) network and the Convolution Neural Network (CNN). BiLSTM makes better use of how text fits into time by looking at both the historical context and the context of what came after.