Berliana Rahmadhani, Purwono Purwono, Alfian Ma'arif, Iswanto Suwarno
Sleep disorders such as insomnia and obstructive sleep apnea (OSA) are frequently underdiagnosed due to the high cost and invasiveness of conventional diagnostic procedures. This study proposes a calibrated stacking ensemble framework for sleep disorder classification using lightweight tabular data. The model integrates LightGBM, CatBoost, and Multi-Layer Perceptron (MLP) as base learners, combined through Logistic Regression as a meta-learner within an out-of-fold scheme. Feature selection was performed using a genetic algorithm, while hyperparameter optimization was conducted using Bayesian optimization via Optuna. Probability calibration was applied using Platt Scaling and Isotonic Regression to improve reliability. The proposed framework was evaluated on two public datasets: Sleep Health and Lifestyle (SHL) and Sleep Disorder Diagnostic (SDD). The model achieved an AUC-macro of 0.9935 and an accuracy of 97.33% on SHL. However, performance decreased on SDD (AUC-macro =0.4656), indicating cross-domain generalization challenges. Compared to the BER-MLP baseline, the proposed method improved accuracy by 2.01% and macro sensitivity by 3.36% on SHL. These findings highlight both the strength of the proposed framework and the importance of domain-aware modeling in clinical machine learning applications. © 2026 IEEE.
Universitas Harapan Bangsa, Department of Informatics, Purwokerto, Indonesia; Universitas Ahmad Dahlan, Departmen of Electrical Engineering, Bantul, Indonesia; Universitas Muhammadiyah Yogyakarta, Departmen of Electrical Engineering, Bantul, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock