A Hybrid CatBoost–ExtraTrees Framework for Workload Variability Classification and QoS Prediction in Multi-Cloud Financial Environments

Abstract

Modern financial systems increasingly rely on multi-cloud environments to manage diverse and dynamic workloads with enhanced scalability, flexibility, and cost efficiency. However, maintaining consistent Quality of Service (QoS) across distributed platforms remains challenging due to fluctuating resource demands, varying service conditions, and strict Service Level Agreement (SLA) constraints. Efficient workload management therefore requires intelligent, adaptive mechanisms that ensure optimal resource utilization while avoiding over- or under-provisioning. To address these challenges, this study proposes a hybrid ensemble framework combining Categorical Boost (CB) and Extra Trees (ET) for workload variability classification and QoS score prediction. The framework adopts a two-stage learning approach integrating regression and classification tasks. In the first stage, regression models predict QoS scores using critical system performance metrics such as CPU utilization, memory usage, latency, throughput, and network bandwidth. In the second stage, the predicted QoS score is incorporated as an additional feature to classify workload variability into Low, Medium, and High categories. The system evaluates multiple machine learning models, including Decision Tree, Random Forest, Gradient Boosting, and AdaBoost, alongside the proposed hybrid CB–ET ensemble. The implementation covers data preprocessing, exploratory data analysis, model training, performance evaluation, and visualization through an interactive Flask-based web interface. Experimental results show that while traditional models achieve moderate performance, the hybrid ensemble significantly improves both regression precision and classification accuracy. By leveraging CB’s capability to model complex feature interactions and ET’s randomness, the framework enhances generalization, robustness, and prediction stability, providing an effective solution for intelligent workload management and QoS optimization in multi-cloud financial environments.