Design and Development of a Hierarchical Confidence Weighted Ensemble Learning Framework for Speech Disorder in Children

Speech and language disorders in pediatric populations can adversely affect communication, learning, cognitive development, and social interaction, making early and accurate diagnosis essential for effective intervention. This study proposes a Hierarchical Confidence-Weighted Ensemble (HCWE) framework for the classification of pediatric therapeutic speech disorders using linguistic and statistical speech-derived features extracted from the CHILDES dataset. The proposed framework integrates multiple calibrated machine learning classifiers, including Random Forest, Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Logistic Regression, within a hierarchical stacking architecture that combines classifier outputs using confidence-weighted probability fusion. To improve predictive performance and generalization, the framework incorporates comprehensive data preprocessing, class balancing using the Synthetic Minority Oversampling Technique (SMOTE), and hybrid feature optimization based on Mutual Information (MI) and Recursive Feature Elimination (RFE). The proposed HCWE was evaluated using standard performance metrics, including accuracy, precision, recall, F1-score, receiver operating characteristic area under the curve (ROC-AUC), calibration analysis, and computational efficiency, while Explainable Artificial Intelligence (XAI) techniques were employed to identify the most influential linguistic features contributing to classification decisions. Experimental results demonstrate that the HCWE framework consistently outperforms individual machine learning models by achieving higher classification accuracy, superior probability calibration, improved robustness, and enhanced predictive reliability in distinguishing Typically Developing (TD) children from those with Specific Language Impairment (SLI). By integrating hierarchical confidence-weighted ensemble learning, linguistic and statistical feature fusion, probability calibration, and explainable AI within a unified framework, the proposed approach provides an accurate, interpretable, and clinically reliable decision-support system for the early identification of pediatric speech and language disorders.