ВЫЯВЛЕНИЕ АНОМАЛИЙ В ДАННЫХ МОНИТОРИНГА ПРОИЗВОДИТЕЛЬНОСТИ С ИСПОЛЬЗОВАНИЕМ АЛГОРИТМА ISOLATION FOREST: ВОЗМОЖНОСТИ И ОГРАНИЧЕНИЯ

This paper explores the application of the Isolation Forest algorithm for detecting anomalies in performance monitoring data of a SaaS project’s servers. The main hypothesis suggests that the algorithm can identify early signs of performance degradation and potential failures by analyzing basic metrics such as CPU load, memory usage, network traffic, and disk space. Two approaches were tested: analyzing each metric separately and aggregating them into a single indicator to assess the overall system state. The results showed that Isolation Forest demonstrates high sensitivity to sudden changes in metrics, leading to a significant number of false positives. This issue is particularly relevant when dealing with short-term metric spikes that do not necessarily indicate real system problems. The paper discusses the limitations of this approach, including the need for fine-tuning hyperparameters, and suggests possible solutions for improving anomaly detection accuracy, such as preprocessing data and combining it with other methods. This study highlights the importance of advanced machine learning techniques in server performance monitoring, especially in conditions with limited metrics, typical of closed-source systems.