Uncovering Irregularities in National Identity Frameworks: An Algorithmic Approach to Anomaly Detection and LLM Based Predictive Auditing

Uncovering Irregularities in National Identity Frameworks: An Algorithmic Approach to Anomaly Detection and LLM Based Predictive Auditing 1Sounak Sarkar, 2Soumyashis Ghosh, 3Sourodeep Kar, 4Dr. Tirthadip Sinha 1,2UG student, Dept. of Computer Science & Engineering, Haldia Institute of Technology, Haldia, West Bengal 3UG student, Dept. of Electronics & Communication Engineering, Haldia Institute of Technology, Haldia, West Bengal 4Assistant Professor, Dept. of Electronics & Communication Engineering, Haldia Institute of Technology, Haldia, West Bengal ABSTRACT National biometric platforms like India's Aadhaar generate massive daily transaction volumes, making traditional reporting too slow to catch operational anomalies or sudden regional demand surges. To shift from reactive auditing to intelligent real-time monitoring, we introduce a predictive analytics engine using real-world UIDAI operational data (March–December 2025). Built on a decoupled React and FastAPI architecture, the platform features an automated time-series forecasting tournament. Using rolling-origin cross-validation with split-conformal prediction intervals, it delivers reliable capacity forecasts with precise statistical bounds. Concurrently, unsupervised Isolation Forests flag localized anomalies across state and district levels without requiring pre-labeled fraud data. A local LLM translates these technical insights into actionable, plain-language briefs backed by a human-in-the-loop audit trail—enabling swift, targeted field responses at national scale.