METHODS AND SYSTEMS FOR CALCULATING NON-PERFORMANCE RISK SCORES BASED ON NON-FINANCIAL TIME SERIES DATA

Abstract

The methodology and system for calculating non-performing risk scores are based on non-financial time-series data collected from two-wheeled vehicle terminals. Characteristic information is inferred from the non-financial time-series data relating to the operation, status, and events of the two-wheeled vehicles. The representation learning module generates latent representation data using a representation learning model trained to learn time-series patterns of usage behavior. The anomaly calculation module computes anomaly scores representing the degree of deviation from normal user behavior from the latent representation data. The scoring module converts anomaly scores into relative rating scores based on a reference distribution to generate non-performing risk scores. The method allows for early risk detection independent of financial data, while reducing false positives and false negatives through relative assessment that takes into account individual and group heterogeneity.

Images