<![CDATA[The rapid proliferation of digital information systems in modern organizations has created an urgent need for proactive performance monitoring and predictive analytics capabilities. Traditional rule-based monitoring approaches are increasingly inadequate in addressing the dynamic, high-dimensional nature of modern computing environments comprising cloud infrastructure, microservices architectures, and distributed databases. This study proposes and evaluates a machine learning (ML)-based framework for predicting and analyzing information system performance metrics in modern computing environments. The framework integrates five supervised and unsupervised ML algorithms — Random Forest (RF), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), Support Vector Machine (SVM), and Isolation Forest — applied to a multi-dimensional dataset of system performance telemetry collected from a heterogeneous IT infrastructure over a 12-month period. The dataset encompasses 14 performance indicators including CPU utilization, memory usage, network throughput, query response time, and application error rates. Experimental results demonstrated that the XGBoost model achieved the highest predictive accuracy (R² = 0.942, RMSE = 2.31%) for CPU load forecasting, while the LSTM network outperformed other models for sequential anomaly detection with F1-score of 0.961. The ensemble approach combining RF and XGBoost reduced false positive rates in performance degradation alerts by 34.7% compared to single-model baselines. The novelty of this research lies in the integration of a hybrid ensemble architecture with SHAP (SHapley Additive exPlanations)-based interpretability analysis, enabling actionable root-cause identification beyond binary anomaly detection — addressing a critical gap in existing AI-based IT performance management solutions.]]>
