AI-Based Autonomous Network Monitoring and Anomaly Detection with Automated Alerts
Detecting anomalous records in network-traffic-like data calls for methods that work without labels and for a way to pass detections to an operator. This paper evaluates a student prototype that combines an Isolation Forest with a reconstruction-based neural detector, merges their binary decisions with a logical OR rule, classifies severity with fixed rules, and builds webhook payloads for an n8n alert workflow. The data are 1,000 synthetic records containing 50 injected anomalies of two kinds. The original implementation scored every record, including records used for fitting, with a fixed reconstruction threshold; re-running it reproduced the saved report exactly, with 286 flagged records and precision 0.175. Under a leakage-free protocol, in which preprocessing, training and thresholds use only a stratified 60% training partition and testing uses an untouched 20% partition, repeated over ten splits, Isolation Forest reached a mean F1-score of 0.82 (precision 0.82, recall 0.85), the reconstruction detector 0.57, and the OR combination 0.75 (recall 0.92, precision 0.64). The OR rule therefore raised recall, mainly for slow-leak-like anomalies, but did not improve the F1-score, and with the original fixed threshold it flagged more than a quarter of normal records. The web interface uses simulated routes, and delivery of alert e-mails was not verified. The results apply to synthetic data only and motivate evaluation on real labelled traffic. Keywords — anomaly detection; Isolation Forest; reconstruction error; autoencoder; OR fusion; automated alerting
Authors
- PONEY JOSHWAA A
- NAVEENESWAR P G
- Mrs. MOGANAPRIYA L
- NITHIYANANDHAM J
Institutions
- E.G.S. Pillay Engineering College (IN)
Publication Details
- Journal
- International Journal of Creative and Open Research in Engineering and Management
- Published
- 2026-11-13
- DOI
- https://doi.org/10.55041/ijcope.v2i11.054
- Primary Topic
- Network Security and Intrusion Detection
- Type
- article
- Field-Weighted Citation Impact
- 0.00