American
International
University-
Bangladesh

Students & Supervisors

Abstract

Research Area:ces Stackomnia, an ensemble learning framework leveraging meta-learning technique for the classification of sleep disorders. Achieving an AUC score of 0.94, the model demonstrates strong predictive performance and highlights the potential of advanced machine learning approaches in addressing the growing prevalence of sleep-related health issues. Objectives Sleep disorders [2], such as insomnia and sleep apnea, significantly impact quality of life and are linked to chronic health conditions. Early prediction is challenging due to heterogeneous symptoms and complex interactions between health factors. This work introduces a stacked ensemble approach to leverage diverse classifiers for robust predictions. The model learns from base classifiers’ outputs and combines them via a meta-learner, addressing limitations of individual models. Methodology The Health and Sleep dataset [1] contains structured information across three primary categories: demographics and lifestyle (e.g., gender, age, occupation), sleep and activity (e.g., sleep duration, sleep quality, physical activity), and health indicators (e.g., stress level, BMI category, blood pressure, and health label). These features enable a holistic analysis of the relationship between lifestyle and health outcomes. A stacked ensemble model was developed using AdaBoost, SVM, and Naive Bayes as base learners, with logistic regression as the meta-learner. Hyperparameter tuning was applied to maximize individual model performance. The final model was evaluated using accuracy, precision, recall, and F1-score to ensure reliable classification across health-related classes. Results and Analysis The stacked ensemble model demonstrated strong classification performance across all evaluated metrics. It achieved an accuracy of 0.9292, with a precision of 0.9285, recall of 0.9292, and an F1-score of 0.9282. These closely aligned values highlight the model’s ability to maintain a consistent balance between identifying relevant instances and minimizing false positives and negatives. Moreover, it can be clearly seen from figure 2 that the model’s ROC curves indicate high discriminative power, with Area Under the Curve (AUC) scores of 0.92 for Class 0, 0.97 for Class 1, and 0.92 for Class 2. These results underscore the model’s robust generalization capability and its effectiveness in distinguishing between multiple sentiment categories."

Keywords

Publication Details

• Type of Publication: Conference 
• Conference Name: IEEE Computer Society Bangladesh Chapter (CS BDC) Summer Symposium 2025
• Date of Conference: 18/07/2025 - 18/07/2025
• Venue: Hajee Mohammad Danesh Science & Technology University, Dinajpur
• Organizer: IEEE Compute society Bangladesh Chapter