Intelligent System for ATM Fraud Detection System using C-LSTM Approach

Abstract

ATMs are vulnerable to a wide variety of assaults and fraud because of the money and personal information available on it. In response, today’s ATMs feature enhanced hardware security systems are capable of identifying specific forms of fraud and manipulation. However, there is no defense in place for future attacks that can’t be anticipated during design. It shows how automated teller machines (ATMs) can be secured against theft without the need for extra hardware. The goal is to employ automatic techniques of model generation to learn normal behavior patterns from the status information of the standard de vices that make up an ATM, with a significant divergence from the taught behavior indicating a fraud attempt. Preprocessing, feature selection, and model training are all parts of the proposed method. Cleaning, integrating, and deduplicating data are all parts of data preprocessing. BOA is employed in feature selection and C-LSTM is used for model training. In C-LSTM, a LSTM recurrent neural network is used to obtain the sentence representation after CNN is used to extract a sequence of higher-level phrase representations. C-LSTM can learn the global and temporal sentence semantics in addition to the local aspects of phrases. When compared to LSTM and CNN, the proposed method fares very well.