<![CDATA[Regression on heterogeneous tabular data remains a challenging problem in machine learning due to mixed numerical and categorical features, non-linear relationships, and context-dependent feature relevance. In many real-world datasets, feature contributions vary across spatial, temporal, and categorical contexts, reducing the effectiveness of conventional regression and ensemble methods that treat all features uniformly. This paper proposes a context-aware ensemble learning framework for regression on heterogeneous tabular data, where contextual information is explicitly modeled through structured feature grouping. Contextual attributes are organized into predefined context groups and integrated into the learning pipeline to capture context-dependent feature interactions. The framework evaluates multiple ensemble models, including Random Forest, XGBoost, and LightGBM, under consistent preprocessing and evaluation settings. Model performance is assessed using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) with k-fold cross-validation, while interpretability is enhanced through Explainable Artificial Intelligence (XAI) techniques using feature importance analysis and SHAP values. Experimental results demonstrate that explicit context modeling consistently improves regression performance across all evaluated ensemble methods compared to baseline approaches. The proposed framework contributes a systematic and generalizable approach to context-aware regression and ensemble interpretability, supported by experimental results showing consistent reductions in MAE and RMSE.]]>
Copyrights © 2026
