Variability-Aware Financial Workload Intelligence via Hybrid Tree Induction and QoS Surface Approximation across Multi-Cloud Substrates

Abstract

Businesses aiming to manage diverse workloads with agility, scalability, and cost efficiency increasingly rely on multi-cloud environments. However, maintaining strict Quality of Service (QoS) requirements across multiple platforms while ensuring seamless resource accessibility remains a significant challenge. Although cloud computing enables the deployment and management of business processes (BPs), handling highly dynamic and fluctuating resource demands in containerized multicloud environments is complex. This often leads to issues such as over-provisioning or underprovisioning of resources, ultimately affecting performance and operational efficiency. Therefore, there is a critical need for intelligent and elastic resource provisioning mechanisms that can satisfy both cloud provider objectives and end-user expectations while adhering to QoS and Service Level Agreement (SLA) constraints. This research proposes a Hybrid Classification and Regression Tree (CART)-based framework to analyze workload variability in multi-cloud environments and accurately predict QoS scores. The system integrates multiple machine learning models, including Decision Tree Classifier (DTC), Random Forest Classifier (RFC), Gradient Boosting Classifier (GBC), and Adaptive Boosting Classifier (AdaBoost). Additionally, a novel hybrid model, termed the Categorical Boosted Extra Trees (CBET), combines the strengths of Categorical Boosting Classifier (CBC) and Extra Trees Classifier (ETC). The framework follows a dual-stage CART pipeline where regression models first estimate QoS scores, which are then used as input features for classification models to categorize workload variability into Low, Medium, and High levels. Experimental results demonstrate that the proposed CBET model significantly outperforms traditional models, achieving higher accuracy, improved generalization, and enhanced robustness.