Consumer Credit Risk Scoring System
Design a machine learning system for a digital lending platform to predict the 12-month probability of default for consumer loan applications. The system must process real-time data from credit bureaus and internal bank transactions, scale to handle 50,000 applications per day with sub-2-second latency, and provide explainable decisioning (reason codes) to comply with financial regulations. Address specific challenges including significant label delay, data drift in changing economic climates, and the sample selection bias inherent in modeling only previously approved applicants.
LightGBMSHAPXGBoostKafkaFlinkSparkMLflowIsotonic RegressionFeature StoreKServe
00