The Implementation of Quantum Annealing for Ensemble Pruning

Ensemble pruning stands as a pivotal strategy in simplifying ensemble complexity while maintaining its accuracy. Although optimization-based methods have historically dominated this field due to their comprehensive solutions, issues with local optima and time-intensive processes persist. Consequently, this study proposes quantum annealing as a quantum-based solution to address these limitations. The approach involves four key stages: formulating pruning problems, embedding formulations into the D-Wave solver, annealing, and reading out the optimal solution. The pruning process aims to maximize ensemble accuracy while ensuring the inclusion of at least one trained learner. The evaluation was conducted across five datasets, focusing on ensemble size and accuracy metrics. A comparative analysis was performed against bagging, AdaBoost, hybrid QA, hybrid QA pruned using particle swarm optimization, and proposed method. The results demonstrate that while the proposed method may not significantly reduce ensemble size compared to particle swarm optimization, it exhibits exceptional performance in all datasets, highlighting its potential to enhance accuracy.