Explainable Data Mining for High-Dimensional Datasets Based on Hybrid Machine Learning Models

Abstract

There are several issues with high-dimensional datasets: feature redundancies, overfitting, computational costs, and the lack of interpretability.While high-dimensional datasets provide rich information for prediction and knowledge discovery, they also pose challenges related to feature redundancy, overfitting, computational costs, and limited interpretability. The hybrid machine learning models combine complementary techniques, including feature selection, dimensionality reduction, ensemble learning, and post hoc explanation, in order to enhance the predictive performance as well as transparency. This paper discusses an integrative review of explainable data mining with hybrid machine learning models in high dimensional datasets. It covers feature selection, random forests, gradient-boosted trees, local surrogate explanations and Shapley-based feature-attribution. The review reveals a major unfilled need: hybrid pipelines tend to focus on the accuracy of their predictions, and less systematic attention is given to the validity, stability, and usefulness of their explanations for human use. The paper brings forward a modular approach which integrates filter and embedded feature selection, ensemble modeling, global and local explanations through SHAP, stability analysis and human-centered validation. Results are qualitative analytical results that are not new experimental results but literature based. The results of the analysis suggest a hybrid approach for interpretable high-dimensional data mining, in which explanations are interpreted as model-dependent evidence—an interpretation that does not mean that explanations are causal evidence. The proposed procedure focuses on minimising leakage, replicability, fidelity to the explanation, and reporting uncertainty transparently.