USING THE ISOLATION FOREST ALGORITHM FOR PROACTIVE ANOMALY DETECTION IN DATA CENTERS

Abstract

As the infrastructure of Data Centers (DCs) becomes increasingly complex, the importance of timely anomaly detection to ensure service continuity is growing. This article investigates the Isolation Forest (iForest) algorithm for the proactive detection of anomalies in multidimensional data streams collected from data center monitoring systems, taking into account both physical and information security. A comparative analysis was conducted with traditional threshold-based approaches and statistical methods.

The experiment was carried out on a synthetic dataset consisting of over 50,000 measurements (CPU, RAM, network traffic, temperature, humidity, power consumption) simulating data from the Zabbix monitoring system. The results showed that the Isolation Forest algorithm achieves high precision (Precision = 0.93), recall (Recall = 0.91), and F1-score (F1 = 0.92) compared to the Local Outlier Factor (LOF) and One-Class SVM methods. The obtained AUC-ROC metric score was 0.94, which proves that the algorithm can operate with high efficiency in a DC environment.