Project / 01 / Mayur Patil
Large-Scale Log Anomaly Detection
Detect unusual HTTP traffic with behavioral features, an anomaly-detection ensemble, and drift monitoring. A separate FastAPI service serves Isolation Forest scores.
The problem
A high volume of server events makes manual inspection impractical. The project explores how to surface unusual behavior and give each alert useful context.
How it works
Separate feature engineering, offline ensemble analysis, and a smaller scoring service. Track experiments with MLflow and inspect feature drift with KS tests and PSI.
Evidence and evaluation
500K synthetic HTTP log records. The deployed Isolation Forest API returned a Critical verdict and a -0.1861 anomaly score for the synthetic DoS preset captured October 14, 2026. This is a serving demonstration, not an accuracy benchmark.
Limitations
The 500K-row dataset is synthetic. Offline ensemble results and the deployed Isolation Forest service are different evaluation targets; they should not be presented as one benchmark.
About the author
Mayur Patil is a data scientist and ML engineer pursuing an M.S. in Data Science at The George Washington University, based in Falls Church, Virginia. His work spans data engineering, applied ML, language models, and sports forecasting.
More about Mayur Patil ↗