A dual pipeline machine learning framework for automated multi class sleep disorder screening using hybrid resampling and ensemble learning
Abstract Accurate classification of sleep disorder related categories, particularly insomnia and sleep apnea, remains an important machine learning research problem. This paper presents a Dual Pipeline Machine Learning Framework for multi class sleep disorder classification using the publicly available Sleep Health and Lifestyle dataset, which is described by its source as synthetic and created for illustrative purposes. The framework evaluates two alternative feature processing strategies: a statistical pipeline utilizing RobustScaler, Mutual Information, and Linear Discriminant Analysis, and a wrapper based pipeline utilizing MinMaxScaler, Boruta feature selection, and an Autoencoder for nonlinear representation learning. SMOTETomek is evaluated to address class imbalance, and nine classifiers are compared through baseline, optimized, and ablation configurations. The dataset was partitioned before learned preprocessing into a 299 sample training set and a permanently held out 75 sample test set. Stratified 8 Fold Cross Validation was used within the training partition for hyperparameter selection, while the final reported metrics were calculated exclusively on the held out test partition. K Nearest Neighbors, XGBoost, LightGBM, and Extra Trees achieved the highest observed test Accuracy of 98.67%, corresponding to 74 correct predictions among 75 test samples, with a 95% Wilson score interval of 92.83% to 99.76%. The results indicate that the effects of normalization, feature selection, and hybrid resampling are strongly classifier dependent. Macro AUC produced a different ranking from accuracy, indicating that single partition accuracy differences of one sample should not be used to order configurations. These findings should be interpreted as methodological evidence from a small synthetic tabular benchmark and do not establish clinical validity, diagnostic reliability, or readiness for real world screening.
Authors
- Md. Sultanul Islam Ovi (ORCID: https://orcid.org/0009-0003-0516-950X)
- Syed Hasan
- Muhsina Tarannum Munfa
- Miftahul Alam Adib
Institutions
- Shahjalal University of Science and Technology (BD)
- George Mason University (US)
- Metropolitan University (BD)
Publication Details
- Journal
- Social Network Analysis and Mining
- Published
- 2026-11-15
- DOI
- https://doi.org/10.1007/s13278-026-01651-y
- Citations
- 1
- Primary Topic
- Imbalanced Data Classification Techniques
- Type
- article
- Field-Weighted Citation Impact
- 1.08