<![CDATA[The development of mobile banking services has increased the convenience of financial transactions, but has also increased the risk of potentially fraudulent transactions. The complexity of transaction patterns and data integrity pose challenges in building an accurate and interpretable detection system. This study aims to analyze the risks of mobile banking transactions using XGBoost and provide an interpretation of factors that influence prediction results using Shapley Additive Explanations (SHAP). The data used are 6,898,889 mobile banking transactions in 2023 with six main attributes. The research stages include data preprocessing, feature extraction based on time and transaction behavior, normalization, data partitioning, anomaly detection using autoencoders, and classification using XGBoost. Autoencoders are used to generate pseudo-labels based on reconstruction error values, while SHAP is used to interpret the contribution of each feature to the model's prediction results. The results show that the autoencoder produces a threshold anomaly of 0.0345 and identifies 1,379 transactions as anomalies. The XGBoost model is able to recognize all anomalous transactions with a recall value of 1.00, although some normal transactions are still predicted as anomalies. The SHAP analysis results show that trx_value is the most influential feature on predictions, followed by channel_type and transaction_hour. This study demonstrates that the combination of Autoencoder, XGBoost, and SHAP can support the detection of potentially fraudulent transactions while providing a more transparent interpretation of the factors influencing the model's decisions.]]>
Copyrights © 2026
