Experimentation of a New Approach Based on Ensemble Learning Estimator to Maximize Accuracy
Lady NLANDU MABUMBI *
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, National Pedagogical University, Democratic Republic of the Congo.
Ornella DIONGA NDIBU
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, National Pedagogical University, Democratic Republic of the Congo.
David-Rissy NKUNGA MBUDI
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, National Pedagogical University, Democratic Republic of the Congo.
Kevin MONGOY BONYOLO
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, National Pedagogical University, Democratic Republic of the Congo.
Chef de Travaux Fader MUKUNA WA MUKUNA
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, National Pedagogical University, Democratic Republic of the Congo.
Pierre KAFUNDA KATALAY
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, University of Kinshasa, Democratic Republic of the Congo.
Joseph KASIAMA NGI ONKOR
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, National Pedagogical University, Democratic Republic of the Congo.
Valery LUKEKA BYEMBA
Centre de Recherche Interdisciplinaire de l'Université Pédagogique Nationale (CRIDUPN), Kinshasa, DRC, Democratic Republic of Congo.
Richard KITONDUA LUBANZDIO
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, National Pedagogical University, Democratic Republic of the Congo.
Cauchy MBAKAS’A KONGOLO
Faculty of Science and Technology, Department of Mathematics, Statistics and Computer Science, National Pedagogical University, Democratic Republic of the Congo.
*Author to whom correspondence should be addressed.
Abstract
In this article, we introduce a novel approach based on an ensemble learning estimator specifically designed to optimize the predictive accuracy of supervised classification models. This contribution, both theoretical and methodological, relies on the strategic combination of multiple heterogeneous learning algorithms (decision trees, boosting methods, SVMs, etc.) orchestrated through a meta-model. The resulting architecture, named MaxEnsForest, is aimed at enhancing the robustness, accuracy, and generalization capacity of traditional models.
Within this framework, we present the results of an extensive series of experiments conducted on several benchmark datasets to evaluate the performance of MaxEnsForest under diverse conditions. The study highlights the individual contributions of each component within the architecture, as well as the impact of integrated optimization strategies such as GridSearchCV, feature importance analysis, and performance visualization through robust evaluation metrics.
This work seeks to establish a rigorous transition from theoretical design to solid experimental validation, empirically demonstrating the relevance and superiority of MaxEnsForest compared to conventional ensemble learning techniques.
Moreover, this research proposes an optimized ensemble learning architecture centered around a Grand Estimator, designed to maximize prediction accuracy while ensuring stability, robustness, and resilience to data variability.
Keywords: Ensemble learning, ensemble estimator, Meta-model, predictive accuracy, generalization performance, hyperparameters, MaxEnsForest