Abstract:
Insomnia happens to be one common sleep disorder that cuts across many
people worldwide. It has left untold suffering in his domains that stretch from
physical to mental to emotional health problems. Polysomnography and clinical
assessment methods are the commonly known traditional ways of diagnosis, but
they happen to be time consuming, resource intensive, and often difficult for
many to access. This study introduces the application of machine learning (ML) for
an automated, efficient, and scalable approach towards insomnia detection. The
physiological and behavioral attributes from students of Daffodil International
University, which are collected into a dataset through pre- processing and analysis,
are used to train various ML models including Gradient Boosting, Random Forest,
and Support Vector Machines. The models were tested against evaluation metrics
like accuracy, precision, recall, and F1- score. The best model is Gradient Boosting,
which achieved testing accuracy of 99.01%, precision of 99%, and F1-score of 99%.
Some of the major challenges addressed in this study are imbalanced datasets,
complex model interpretability, and ethical considerations such as data privacy.
By these results, machine learning turned out to be a feasible option for early
detection and accurate latticing of insomnia. Such a system has the potential to
revolutionize healthcare by providing accessible, noninvasive, and cost-effective
diagnostic tools, thereby improving patient outcomes and advancing the role of ML
in sleep medicine.
