Sleep disorders pose a significant challenge to worldwide health, underscoring the critical demand for accurate and prompt diagnostic methods. This study explores the use of machine learning (ML) methods to enhance and automate diagnostic procedures in healthcare systems for treating sleep disorders. A comprehensive dataset of physiological and behavioral sleep-related attributes was analyzed to evaluate and compare the performance of multiple ML algorithms, including Naive Bayes, Linear Discriminant Analysis (LDA), XGBoost Classifier, Gradient Boost Classifier etc. These models were evaluated with important metrics including accuracy, precision, recall, and F1score, and cross-validation was used to maintain reliability and strength. The analysis also considered computational efficiency and model complexity. Data preprocessing involved addressing missing values, feature scaling, and exploratory data analysis, with additional optimization through parameter tuning and feature selection. Notably, the KNN model was further optimized using the Fish Swarm Optimization technique, achieving an improved accuracy of 95.56%, surpassing its initial performance of 94.25%. This optimization underscores the novelty of the study, highlighting the potential of hybrid approaches in advancing MLdriven healthcare diagnostics.