Explainable AI-Based Performance Prediction Using Optimised ML Models in Cloud Systems

Abstract

Performance prediction by accurately estimating the behaviour in cloud computing systems is essential for improving resource utilisation and operational predictability. This work presents performance classification of cloud tasks using low-level infrastructure metrics and optimised machine learning (ML) models - XGBoost, Support Vector Machines (SVM), and Random Forest - with data collected using SmartMonit, a real-time monitoring system. A heavy preprocessing pipeline was applied to the dataset, consisting of outlier detection, time-aware grouping, and min-max normalisation. Model effectiveness was enhanced through hyperparameter tuning with GridSearchCV. Following training, the models were interpreted using Explainable Artificial Intelligence (XAI) techniques such as SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations), which allowed for post hoc analysis of model decisions and feature importances without altering the model development pipeline. In addition, a genetic algorithm-based feature weight optimisation strategy was applied using SHAP-derived importance scores, allowing explanation-guided reweighting to enhance predictive performance. The optimised models achieved accuracy scores above 92%, corresponding to a relative improvement of 3-4% over the baseline, indicating that well-tuned ML models with XAI and explanation-driven optimisation can contribute to efficient transparency and decision-making support in dynamic clouds. © 2025 IEEE.