An Adaptive Fraud Detection Algorithm in Online Transaction based on CNN-HSVM Approach

Online transaction fraud is hard to detect due of users' diverse transaction behaviours. Using the same criterion for all users could lead to erroneous ratings and reduce fraud detection. To improve precision, a transactional behaviorbased approach is needed. This research proposes detecting fraudulent transactions using a hypersphere model that examines user-specific activities. Based on the user's historical transaction patterns, the model creates multidimensional transaction behaviour traits. Preprocessing, feature extraction, and model training comprise the detection strategy. Data preparation involves cleaning, assessing, and processing incomplete, biassed, or noisy data. Feature extraction uses supervised data mining to uncover pre-fraud buying habits. Training a CNN-HVSVM model to detect fraudulent transactions follows. The proposed approach outperforms HVSVM and CNN if applied. The model improved accuracy to 96.73% detection. A hypersphere model-based approach to online transaction fraud detection can improve accuracy by leveraging individual transaction behaviours, according to this study. Fraud prediction is now more accurate due to the methodology's quick examination of historical transaction data and attribute identification.