Development of a Stochastic Gradient Descent-Based Neuro Model For Identifying Fraudulent Transactions

Abstract

The increase in cases of fraudulent transactions around the world has necessitated the exploration of artificial intelligence techniques to address this menace. This paper explored the powerful matching properties of artificial neural networks for data analysis and modeling of financial transactions. An artificial neural network model was optimized, trained with a gradient descent algorithm, and deployed for modeling and analyzing fraudulent transactions under different conditions. Performance results obtained using this algorithm were cross-analyzed with those obtained with the same network using other algorithms. With a Variance Acount-For (VAF) of 98.64 for training and 96.27 for testing, together with the advantage of being fast, accurate, and efficient, our findings show that the performance of the neuro model is satisfactory and can be viewed as a promising technique in the domain of financial security and information forensics.