Anshita Vats

AN AUTOMATED MACHINE LEARNING FRAMEWORK FOR DETECTING ANTI_FOREENSIC INVESTIGATIONS

Achieved 98.3% accuracy using Random Forest on a 10,000-file balanced dataset Achieved 98.1% accuracy using ANN — second best performing model Classified 3 file classes (Normal, Encrypted, Steganography) simultaneously in a single framework — first study of its kind Covered 6 file formats (DOC, PDF, PPT, CSV, TXT, JPG) in one unified model Extracted 9 forensic features from raw binary data with zero content access — preserving legal chain of custody Validated with 5-fold stratified cross-validation — confirmed no result was due to lucky data split Reduced investigator workload by enabling automated triage of large evidence sets 100% precision and recall on Encrypted class — no encrypted file was missed or misclassified Framework processes files without decryption keys — legally safe for court use Dataset generated ethically — no real sensitive data used at any point

AI-Powered Crowd Management and Risk Alert System Using Video Analytics

Delivered real-time crowd-density detection at 30–60 FPS, validated across 7–8 test video sequences under varying density conditions. Designed a context-adaptive risk model classifying density as Safe, Moderate, or Dangerous (3.0–5.0 persons/m²) across a 7-stage detection pipeline with 4-zone spatial partitioning. Incorporated behavioral anomaly detection and a 16-field audit log for forensic traceability, while benchmarking AI surveillance adoption across 21 Indian states. Built entirely on open-source tools, keeping false-positive rates under 15% for reliable operational use.