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

Md Sultanul Islam Ovi, Muhsina Tarannum Munfa, Miftahul Alam Adib, Syed Sabbir Hasan · 2026

Read A Dual Pipeline Machine Learning Framework for Automated Multi Class Sleep Disorder Screening Using Hybrid Resampling and Ensemble Learning by Md Sultanul Islam Ovi, Muhsina Tarannum Munfa, Miftahul Alam Adib, Syed Sabbir Hasan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-05814.txt.

Introduction

Md Sultanul Islam Ovi and 3 co-authors present in 2026 a dual-pipeline machine learning approach for screening sleep disorders such as insomnia and sleep apnea. Clinical sleep studies are accurate but expensive and hard to scale for population screening. Their system works from a sleep health and lifestyle dataset instead.

One pipeline targets linear patterns using mutual information and discriminant analysis. The other selects aspects with the Boruta method and learns nonlinear representations with an autoencoder. A hybrid resampling plan corrects class imbalance between disorder categories. Extra Trees and nearest-neighbor classifiers reportedly reached the best accuracy, beating recent baselines on the same data, with statistical tests backing the gain and fast inference suitable for screening use.

Catalog record

CREATORMd Sultanul Islam Ovi, Muhsina Tarannum Munfa, Miftahul Alam Adib, Syed Sabbir Hasan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-05814

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Dual Pipeline Machine Learning Framework for Automated Multi Class Sleep Disorder Screening Using Hybrid Resampling and Ensemble Learning Md Sultanul Islam Ovi, Muhsina Tarannum Munfa, Miftahul Alam Adib, Syed Sabbir Hasan Abstract This manuscript is a preprint and is currently under review for journal publication. 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. Keywords: Sleep Disorder Screening, Dual Pipeline Framework, SMOTETomek, Feature Engineering, Ensemble Learning 1 Introduction Sleep disorders constitute a pervasive global health challenge, severely diminishing cognitive function, physical health, and overall quality of life. Among the most deleterious of these conditions are Insomnia and Obstructive Sleep Apnea (OSA), which increasingly affect the young adult population Ahadian et al. 2024 . Insomnia is characterized by chronic difficulties in sleep initiation or maintenance, leading to daytime fatigue and significant cognitive alterations Alshammari 2024 . The chronic persistence of these disorders exacerbates the risk of severe comorbidities, including cardiovascular disease and neurocognitive degeneration. Consequently, the accurate and timely diagnosis of emerging sleep pathologies is critical. However, traditional diagnostic modalities, particularly the gold standard polysomnography (PSG), are resource intensive, necessitate expert invigilation, and impose significant temporal constraints. These limitations render PSG inaccessible for long duration monitoring and impractical for large scale population screening Sarala et al. 2025 . The epidemiological scale of this issue is substantial. Recent reports indicate that OSA alone affects approximately one billion adults aged 30 to 69 worldwide Benjafield et al. 2019 . Similarly, insomnia prevalence has surged, affecting nearly 20% of young adults who experience intermittent symptoms. If left unaddressed, the chronic sleep deficits resulting from these disorders compound cardiometabolic and neurocognitive risks, creating a pressing public health burden Morin and Jarrin 2022 . To address the scalability limitations of clinical PSG, Machine Learning (ML) has emerged as a promising paradigm for automated disorder screening. Recent advancements in algorithmic efficacy allow for the analysis of complex physiological signals, demographic profiles, and lifestyle factors to predict diagnostic risks with high precision. Ensemble approaches, deep learning neural networks, and hybrid models have demonstrated the capacity to estimate risk accurately Rahman et al. 2025 . Furthermore, the integration of these models with modern wearable sensors facilitates cost effective, home based monitoring, thereby mitigating the data acquisition bottlenecks inherent in conventional clinical environments Monowar et al. 2025 ; Imtiaz 2021 . ML based solutions utilizing limited input modalities, such as nocturnal oximetry or photoplethysmography (PPG), have opened new opportunities for risk stratification even when the patient is at home Mashaqi et al. 2020 . Previous studies employing Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), such as the DeepSleepNet architecture, have shown success in staging from single channel EEG, while U time approaches have excelled in segmentation tasks Supratak et al. 2017 . Despite these advancements, existing ML based solutions often struggle with the heterogeneous nature of lifestyle data and severe class imbalance, which can bias predictions toward healthy classes. While recent benchmarks by Rahman et al. Rahman et al. 2025 and Monowar et al. Monowar et al. 2025 have achieved high accuracy, there remains a need for a framework that rigorously handles non linear feature interactions and ensures statistical robustness in its predictions. In this study, we propose a Dual Pipeline Machine Learning Framework designed to maximize diagnostic precision for multi class sleep disorder screening. By integrating rigorous statistical feature engineering with advanced wrapper based selection methods, our approach addresses the limitations of prior works. The primary contributions of this paper are as follows: 1. We introduce a novel Dual Pipeline architecture that processes data through two parallel paths: a statistical pipeline focusing on linear separability via Mutual Information and LDA, and a wrapper based pipeline leveraging Boruta and Autoencoders for non linear feature extraction. 2. We implement a hybrid class balancing strategy using SMOTETomek, which simultaneously oversamples minority classes and cleans class boundaries to mitigate bias against Insomnia and Sleep Apnea cases. 3. We achieve state of the art classification performance using Extra Trees and K Nearest Neighbors classifiers, outperforming recent 2025 benchmarks on the Sleep Health and Lifestyle dataset. 4. We provide a rigorous statistical validation using the Wilcoxon Signed Rank Test to confirm that the performance improvements are statistically significant and not artifacts of random variance. The remainder of this paper is organized as follows. Section 2 reviews the relevant literature and identifies existing research gaps. Section 3 details the proposed methodology, including data preprocessing, the dual pipeline architecture, and the classification algorithms employed. Section 4 presents the experimental results, ablation studies, and a comprehensive discussion of the findings. Section 5 addresses the ethical considerations inherent in deploying AI for healthcare screening. Finally, Section 6 provides the conclusion and outlines directions for future research. 2 Literature Review The early detection of sleep disorders such as Sleep Apnea and Insomnia is critical for mitigating long term cardiovascular and neurocognitive risks. While traditional polysomnography remains the gold standard, its resource intensity has driven a paradigm shift toward Machine Learning (ML) based screening. This review critically analyzes the evolution of these automated approaches, categorizing them into signal based and lifestyle based modalities, to identify the methodological gaps that necessitate the proposed Dual Pipeline Framework. 2.1 Signal Based Detection Methodologies The majority of existing literature focuses on detecting disorders, particularly Obstructive Sleep Apnea (OSA), using physiological signals like ECG or SpO2. Kristiansen et al. Kristiansen et al. 2021 and Bahrami et al. Bahrami and Forouzanfar 2022 demonstrated that deep learning architectures could achieve accuracies between 85% and 89% using multi signal inputs. Similarly, Serrano Alarcón et al. Alarcon et al. 2023 utilized 1D CNNs on portable monitor signals to detect events with 84.3% accuracy. While effective, these approaches inherently require physical sensors and continuous monitoring equipment, limiting their scalability for mass population screening or non contact scenarios. Furthermore, studies like Khandelwal et al. Khandelwal et al. 2023 acknowledge that reducing data dimensionality remains a challenge in these high frequency signal domains. In the context of Insomnia, researchers such as Tripathi et al. Tripathi et al. 2022 and Sharma et al. Sharma et al. 2021 successfully leveraged Heart Rate Variability (HRV) features, achieving accuracies exceeding 97% with ensemble and KNN classifiers. However, these models rely heavily on the

Read the complete text