Predictive capabilities analysis for fault detection in critical IT services: case study of ASAP BUSINESS S.A.

Abstract

This scientific article explores the implementation of machine learning models to predict CPU and RAM congestion in virtual environments using Zabbix for monitoring and Grafana for visualization. A controlled testing environment, hosted on servers located in Germany, simulates high-load conditions through the stress-ng tool, generating datasets with temporal and usage variables. The data is cleaned, structured, and processed in Python for binary classification using Random Forest, Support Vector Machine, and K-Nearest Neighbors. The models aim to identify congestion events where resource usage exceeds 85%. Evaluation metrics such as accuracy, precision, recall, and F1-score confirm that Random Forest delivers superior performance across all categories. The study highlights the effectiveness of time-based variables and demonstrates that predictive monitoring improves operational stability. It concludes that integrating artificial intelligence into existing monitoring systems enables proactive responses to potential failures, optimizes resource usage, and enhances service continuity. The research offers a replicable, low-cost methodology applicable to various IT infrastructures.