Enhancing Multi-Tenant Database Resource Allocation with Active Learning-Driven Multi-Objective Optimization

We present an enhanced approach for MultiTenant Database Resource Management Systems (MTDRMS) that integrates Active Learning (AL) with MultiObjective Optimization Algorithms (MOOA). Our methodology dynamically identifies and prioritizes the most valuable data points during optimization, significantly reducing computational demands while maintaining or improving allocation quality. The AL framework employs an iterative query strategy to systematically evaluate highpotential data points that offer maximal optimization gains. A Transformer-based architecture further enhances this process by accurately predicting data point informativeness, enabling effective scaling across large multi-tenant deployments. This integrated solution operates through coordinated data collection, intelligent query selection, optimization execution, and continuous model refinement collectively optimizing the resource allocation pipeline. Experimental validation shows consistent improvements in both convergence speed and decision quality, with particular advantages in complex, high-dimensional optimization scenarios characteristic of modern database systems.