Authors:
G. Srimayi, Siri Amara, C. Shakila, P. Divya, Sai Vishaal Saibaskar

Addresses:
Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Computer Science, Veltech Ranga Sanku Arts College, Thiruvallur, Tamil Nadu, India. Department of Computer Science and Engineering, Dhaanish Ahmed College of Engineering, Chennai, Tamil Nadu, India. Department of Data Science, University of Wisconsin, Madison, Wisconsin, United States of America.

Abstract:

Transparent normal living data benchmarks and neural networks analyze sleep quality. Lab-based sleep studies and the Pittsburgh Sleep Quality Index (PSQI) are the gold standard for sleep quality evaluation. Still, many educational and everyday settings need a cheaper, simpler method that doesn’t require specialized instruments. A neural network trained on age, sleep duration, physical activity, stress, body-mass category, occupation, heart rate, daily steps, and blood pressure from a public lifestyle benchmark predicts sleep quality in this unique, transparent, and reproducible experiment. A claims benchmark tabular dataset, not a clinical cohort. To prevent the algorithm from memorizing repetitive patterns, remove redundant feature profiles before separating 374 rows into 109 profiles. Simple MLP classifier with 24-dimensional encoded input, 128 and 64-unit hidden layers, ReLU activations, Adam optimization, early pausing, and split-only class-balancing. To compete with logistic regression, RBF SVM, and random forest, transparent preprocessing uses one-hot encoding, blood-pressure decomposition, numerical scaling, and 3-fold stratified cross-validation. On the held-out test set, the upgraded neural model had 0.773 accuracy, 0.739 macro F1, 0.364 mean absolute error, 0.864 within-1-score accuracy, and 0.717 mean predictive confidence. Permutation testing showed that stress and sleep length matter most. Random forest outperformed the neural model in the tabular test, with a large difference. Lifestyle-based sleep screening is affordable because the neural network passed class-sensitive testing. Repeatable method, URL-verified references, prototype interface appendices.

Keywords: Sleep-Quality Prediction; Neural Networks; Lifestyle Analytics; Tabular Machine Learning; Physical Activity; Sleep Duration; Reproducible Research; ReLU Activations.

Received on: 25/05/2025, Revised on: 28/07/2025, Accepted on: 13/09/2025, Published on: 03/03/2026

DOI: 10.69888/FTSNL.2026.000644

FMDB Transactions on Sustainable Neuroscience Letters, 2026 Vol. 1 No. 1, Pages: 48–59

• Views : 31
• Downloads : 9

Download PDF