AutoDQ: Adaptive Data Quality Control Based on Distribution Drift and Historical Anomalies

Modern data quality systems commonly rely on static validation rules and manually updated thresholds. This design is fragile in non-stationary data streams, where normal sensor behavior, market regimes, and clinical populations evolve over time. This paper presents AutoDQ, an adaptive methodology for automated data quality control. AutoDQ combines data profiling, baseline validation checks, multi-method distribution drift detection, ensemble anomaly detection, adaptive threshold reconfiguration, and an integral risk score. Experiments on synthetic IoT, fintech, and medical datasets show that the full AutoDQ configuration eliminates false column-level alarms on clean data in the tested benchmark, while baseline configurations produce substantially higher false alarm rates. The results indicate that closed-loop adaptive data quality monitoring can reduce alarm fatigue while preserving sensitivity to real drift and anomalies.