Research on Comprehensive Evaluation of Ensemble Learning Based on RTOPSIS
Stacking ensemble learning is often considered a"black-box"model,utilizing predictions from multiple base learners as input and employing a meta-learner to generate the final prediction.This complexity makes it difficult to understand exactly how each base learner contributes to the final result.To address this issue,we propose the RTOPSIS method.This method combines grey relational co-efficient calculations with the TOPSIS method,providing decision-makers with an effective tool to clearly reveal the contribution of each base learner in the Stacking model to the final outcome.Specifically,the RTOPSIS algorithm is employed as a substitute for traditional discriminative methods,comprehensively considering the relationships between base learners and meta-learners to yield more objective and rational model ranking results;The grey relational analysis algorithm is applied to calculate the weights of individual base learners in the Stacking model and reflect their contributions to the final outcome.Experimental results demonstrate that in the comprehensive evaluation of Stacking models,relative to singular metrics such as Accuracy,AUC,and F1-score,the RTOPSIS algorithm provides more appropriate rankings for the six models considered in this study,and is largely consistent with the ranking results of the classical distance-based TOPSIS algorithm.Therefore,the proposed RTOPSIS algorithm exhibits a more comprehensive evaluation effect in Stacking model assessment.
technique for order preference by similarity to an ideal solutiongrey correlation degreeweightStacking modelcomprehensive evaluation criteria
国家科技期刊平台
NSTL
万方数据
