A bill abnormality detection and difference analysis method based on time series prediction and isolated forest
By fusing trend residual signals from time-series prediction and isolated forest models in bill anomaly detection, and combining dynamic thresholds and anomaly fingerprint databases, the problems of feature redundancy and high false alarm rates in existing technologies are solved, achieving efficient bill anomaly detection and discrepancy analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Filing Date
- 2026-05-20
- Publication Date
- 2026-07-17
AI Technical Summary
The existing dual-engine parallel detection architecture suffers from problems such as feature redundancy, high false alarm rate, high operation and maintenance cost, and disconnect between detection and analysis. In particular, the false alarm rate increases during business peaks and settlement days, leading to an increase in manual review work orders.
By explicitly injecting the trend residual signal of the time series prediction model into the isolated forest model, and using a dynamic threshold adjustment mechanism, a unified feature wide table and anomaly fingerprint database are constructed to realize model feature fusion and automated analysis path, reducing redundant calculations and manual intervention.
It significantly reduced the false alarm rate, improved the ability to identify trend anomalies, reduced operating costs, accelerated the efficiency of difference analysis, and enabled a direct jump from anomaly detection to root cause localization.
Smart Images
Figure CN122413245A_ABST