DMSPC: A novel approach for meta learner selection in stacked ensemble to enhance speech emotion recognition

Speech emotion recognition plays a vital role in human–computer interaction. In this research work, we propose a Dynamic Meta-Selection approach in Stacked Ensemble. The proposed model is structured with two distinct layers. In the first layer we select the datasets, perform pre-processing and extract handcraft features. In the second layer we selected different machine learning classifiers, as base learners. Further, we introduce a novel meta-selection approach DMSPC, which dynamically selects the most suitable base classifier as the meta-classifier. To validate the results we have taken into account four distinct datasets: SUBESCO, RAVDESS, EMOVO, and EMODB. The results demonstrate that the proposed model achieves promising accuracy across multiple datasets – 79.39% on RAVDESS, 91% on SUBESCO, 84.44% on EMOVO, and 84.55% on EMO-DB – showing superior performance when compared to various traditional ML classifiers.