A Deep Neuro-Adaptive Synthetic Model for the Detection of E-Commerce Fraud Transactions

Fraud detection is a critical aspect of safeguarding financial systems and online transactions. Traditional methods often face challenges in handling imbalanced datasets, where fraudulent instances are significantly outnumbered by legitimate transactions. This research explores the effectiveness of combining deep learning methods with Adaptive Synthetic Sampling (ADASYN) to improve the performance of fraud detection models. Experimentation on python shows that the proposed DNN with ADASYN model achieved the best and highest classification accuracy of 97.8% as against the existing algorithm including DT with SMOTE which achieved 91%, NB with SMOTE which achieved 95% and RF with SMOTE which achieved 95% respectively. thus, from the experiment, it is noticed that addressing data class imbalance using techniques like ADASYN and SMOTE can positively impact fraud detection accuracy by mitigating the challenges posed by imbalanced datasets. The successful development of the proposed method has extended the detection accuracy, precision, recall and F-score of the methods compared to other classical machine learning methods. Thus, this  enhances the effective fraud detection system for e- commerce security and trustworthiness of the platform  protect users from fraudulent activities, reduce financial losses, and preserve the platform's reputation.

Keywords:- Deep Neural Network; Feature Extraction; Machine Learning; Fraud Detection, ADASYN.