CCNN: An Artificial Intelligent based Classifier to Credit Card Fraud Detection System with Optimized Cognitive Learning Model

Abstract

Nowadays digital transactions play a vital role in money transaction processes. Last 5 years statistical report portrays the growth of internet money transaction especially credit card and unified payments interface. Mean time increasing numerous banking threats and digital transaction fraud rates also growing significantly. Data engineering techniques provide ultra supports to detect credit card forgery problems in online and offline mode transactions. This credit card fraud detection (CCFD) and prevention-based data processing issues raising because of two major reasons first, classification rate of legitimate and forgery uses is frequently changing, and next one is fraud detection dataset values are vastly asymmetric. Through this research work investigating performance of various existing classifier with our proposed cognitive convolutional neural network (CCNN) classifier. Existing classifiers like Logistic Regression (LR), K-nearest neighbor (KNN), Decision Tree (DT) and Support Vector Machine (SVM). These models are facing various challenges of low performance rate and high complexity because of low hit rate and accuracy. Through this research work we introduce cognitive learning-based CCNN classifier methodology with artificial intelligence for achieve maximum accuracy rate and minimal complexity issues. For experimental data analysis uses dataset of credit card transactions attained from specific region cardholders containing 284500 transactions and its various features. Also, this dataset contains unstructured and non-dimensional data are converted into structured data with the help of over sample and under sample method. Performance analysis shows proposed CCNN classifier model provide significant improvement on accuracy, specificity, sensitivity and hit rate. The results are shown in comparison. After cross-validation, the accuracy of the CCNN classification algorithm model for transaction fraudulent detection archived 99% which using the over-sampling model.