2026-10-12T03:41:18Z https://data.mendeley.com/oai

oai:data.mendeley.com/3v5bsvxdfz.1 2026-09-22T16:54:18Z

Vairal, Kushal AI-powered-Log-Anomaly-Detection-System Mendeley Data Log files are the primary source of runtime telemetry for software applications, but their volume, speed, and unstructured nature make manual inspection or static rule-based monitoring (such as regular expressions or grep-based error filtering) highly impractical. Traditional methods cannot capture context exceptions such as gradual performance degradation or exception errors. It has patterns and silent security holes and requires constant manual rule maintenance. This paper presents a real-time log anomaly detection system that bridges the gap between traditional monitoring and autonomous detection. The system integrates a Java Spring Boot backend for log processing, parsing, normalization, and persistence with a lightweight Python Flask microservice that serves an unsupervised Isolation forest model. Seven engineered features are extracted from each log entry: severity level, message length, number of error keywords, number of warning keywords, presence of stack traces, time of day, and weekend activities. In addition to marking outliers, the system normalizes raw Isolation Forest decision scores into a [0,1] probability range and generates textual reasoning for each anomaly to address the black-box explain ability gap of both deep learning based and LLM-based anomaly detectors. When tested on real-world workloads, the system enables low latency detection, online retraining, and provides real time alerts displayed on an interactive dashboard Machine Learning Log Analysis Real Time Optimization Dataset doi:10.17632/3v5bsvxdfz.1 oai:data.mendeley.com/3v5bsvxdfz.1 Creative Commons Attribution 4.0 International http://creativecommons.org/licenses/by/4.0 https://data.mendeley.com/datasets/3v5bsvxdfz 2026-09-22T16:54:18Z