Volume 15, Number 1/2

Lightweight ML-Based System Classification for Resource-Constrained Monitoring Environments

  Authors

Jeremy D. Howard, Independent Scholar, USA

  Abstract

Modern system monitoring in distributed and resource-constrained environments is limited by reduced observability, high-overhead instrumentation, and reliance on rule-based or computationally intensive machine learning models. This study presents a lightweight machine learning pipeline for system-state classification using structured synthetic telemetry data, enabling non-intrusive deployment without reliance on sensitive operational inputs. Logistic Regression achieved an accuracy of 0.97, with precision of 0.973, recall of 0.947, and F1-score of 0.959, outperforming a Random Forest model (F1 = 0.909). Five-fold cross-validation yielded a mean F1-score of 0.980 (±0.012), indicating stable performance across data partitions. A rule-based baseline failed to detect degraded states (F1 = 0.0), highlighting the limitations of static threshold-based monitoring. Average inference time was approximately 0.006 ms per sample under local conditions. These findings demonstrate that low-complexity, interpretable models can provide reliable and efficient system-state classification in constrained environments.

  Keywords

Lightweight Machine Learning, System Classification, & Resource-Constrained Environment