STUNTING CLASSIFICATION IN CHILDREN USING VIOLA-JONES AND MULTI-FEATURE FUSION WITH PRE-TRAINED MODELS

Abstract

Stunting remains a critical public health issue, particularly in developing countries, where early detection plays a vital role in prevention and intervention. Previous studies have generally relied on single-feature approaches, either using handcrafted descriptors or convolutional neural networks (CNNs) alone, which often fail to capture subtle craniofacial differences associated with stunting. This study proposes an image-based classification system for detecting stunting in children using facial analysis. The proposed method integrates Viola–Jones face detection with facial landmarks, Gray Level Co-occurrence Matrix (GLCM), Color Co-occurrence Matrix (CCM), and local descriptors such as SIFT–FAST/ORB, combined with deep features extracted from a pre-trained EfficientNet model. Feature fusion was performed by concatenating handcrafted and deep features before classification using a fully connected layer with Softmax activation. Experimental results demonstrated that the proposed fusion model achieved superior performance compared to single-feature baselines, reaching 98% accuracy, 0.98 precision, 0.97 recall, and an F1-score of 0.98. These findings indicate that the integration of geometric, texture, color, and deep semantic cues effectively enhances sensitivity toward the stunting class and improves model interpretability. The novelty of this study lies in the combination of classical computer vision and deep learning techniques for robust, interpretable, and clinically relevant stunting detection. This approach offers strong potential for developing digital health tools that enable early, non-invasive stunting screening in children.