In response to the diverse evolution of fintech crimes—manifesting in behavioral pathways, identity spoofing, and cross-platform attacks—this study investigates a multi-stage risk identification and early warning model integrating big data analytics. The model achieves hierarchical identification of anomalous accounts through structural feature screening, deep modeling of temporal behaviors, and a Stacking ensemble strategy. It further incorporates a dynamic threshold mechanism and an online self-learning module to enhance adaptability. A testing platform under real business conditions was constructed. Comparisons with models such as GBDT, BiLSTM, and Transformer showed that the proposed model achieved an AUC of 0.941 and an F1-score of 0.893, with a significantly lower standard deviation than other methods, demonstrating strong stability and robustness.
Li et al. (Thu,) studied this question.