Trisnani Widowati, Ade Novi Nurul Ihsani, Anik Maghfiroh, Clarita Aprilliani, Septian Eko Prasetyo
Hair loss is a multifactorial condition that requires accurate classification to support reliable and personalized decision-making. However, many existing machine learning approaches suffer from data leakage and limited interpretability, reducing their robustness and practical applicability in decision support systems. This study proposes a leakage-resistant machine learning framework for multi-class hair loss classification, integrating explainable artificial intelligence to enhance transparency and reliability. The framework employs a unified preprocessing pipeline within nested cross-validation to prevent information leakage, while SMOTEENN is used to address class imbalance. Several algorithms, including Decision Tree, Random Forest, K-Nearest Neighbors, Logistic Regression, and Extreme Gradient Boosting, are evaluated within this pipeline. Experimental results indicate that Extreme Gradient Boosting achieves the best performance, with an accuracy of 0.8300, F1-score of 0.7908, and AUC of 0.9305 in nested cross-validation. Evaluation on a holdout dataset demonstrates stable generalization, achieving an accuracy of 0.8250, F1-score of 0.7517, and AUC of 0.9182. Furthermore, the integration of explainable artificial intelligence enables interpretable predictions that can be utilized in a rule-based decision support system. Overall, the proposed framework provides a robust, transparent, and leakage-resistant solution for reliable machine learning-based classification in practical applications. © International Journal of Technology in Education and Science (IJTES). This is an open access article under the CC BY-NC-SA license (http://creativecommons.org/licenses/by-nc-sa/4.0/).
Beauty Education Study Program, Faculty of Engineering, Universitas Negeri Semarang, Indonesia; Indonesia; Department of Electrical Engineering, Faculty of Engineering, Universitas Negeri Semarang, Indonesia
