A Dual Pipeline Machine Learning Framework for Automated Multi Class Sleep Disorder Screening Using Hybrid Resampling and Ensemble Learning

TL;DR

This work tackles the challenge of scalable, non-invasive sleep disorder screening by introducing a Dual Pipeline Machine Learning Framework that processes lifestyle- and physiology-based data. It splits feature engineering into two parallel paths—statistical (Mutual Information + Linear Discriminant Analysis) and wrapper-based (Boruta + Autoencoder)—and uses SMOTETomek to balance classes, achieving state-of-the-art accuracy (up to 98.67%) on the Sleep Health and Lifestyle dataset with sub-400 ms inference. The approach is validated with rigorous cross-validation and Wilcoxon Signed Rank Testing, demonstrating statistically significant improvements over baselines. The framework offers practical implications for population-scale screening and potential integration with wearables, while acknowledging limitations such as dataset size and reliance on self-reported lifestyle metrics.

Abstract

Accurate classification of sleep disorders, particularly insomnia and sleep apnea, is important for reducing long term health risks and improving patient quality of life. However, clinical sleep studies are resource intensive and are difficult to scale for population level screening. This paper presents a Dual Pipeline Machine Learning Framework for multi class sleep disorder screening using the Sleep Health and Lifestyle dataset. The framework consists of two parallel processing streams: a statistical pipeline that targets linear separability using Mutual Information and Linear Discriminant Analysis, and a wrapper based pipeline that applies Boruta feature selection with an autoencoder for non linear representation learning. To address class imbalance, we use the hybrid SMOTETomek resampling strategy. In experiments, Extra Trees and K Nearest Neighbors achieved an accuracy of 98.67%, outperforming recent baselines on the same dataset. Statistical testing using the Wilcoxon Signed Rank Test indicates that the improvement over baseline configurations is significant, and inference latency remains below 400 milliseconds. These results suggest that the proposed dual pipeline design supports accurate and efficient automated screening for non invasive sleep disorder risk stratification.

Figures (5)

• Figure 1: Schematic representation of the proposed Dual-Pipeline Machine Learning Framework, illustrating the parallel processing paths for statistical (Pipeline 1) and wrapper-based (Pipeline 2) feature engineering.
• Figure 2: Feature Importance Analysis illustrating the relative predictive power of the original and engineered features derived from tree-based ensemble estimators.
• Figure 3: Visualization of the feature space distribution illustrating the transition from the imbalanced original dataset to the balanced dataset following SMOTETomek application.
• Figure 4: Schematic overview of Pipeline 1 showing the sequential application of outlier-robust scaling, mutual information-based selection, and linear projection via LDA to optimize the feature space for linear separability.
• Figure 5: Schematic overview of Pipeline 2 illustrating the workflow of range-based normalization, all-relevant feature selection using Boruta, and non-linear dimensionality reduction via an Autoencoder network.