<![CDATA[Sleep disorders are health problems that can affect quality of life and have the potential to increase the risk of various chronic diseases. Therefore, a computational approach is needed to accurately and efficiently classify sleep disorders. The ANN model used has a two-layer hidden architecture with 128 and 64 neurons, respectively, and uses the ReLU activation function, equipped with a dropout layer to reduce overfitting. Three neurons with a softmax activation function make up the output layer, which produces probabilities for every class. To improve model performance, three feature selection methods were compared, namely Chi-Square, Information Gain, and Pearson Correlation. The test results showed that the ANN model without feature selection produced an accuracy of 89.3%. After feature selection, the model's performance improved significantly. The Chi-Square method produced 8 selected features with the highest accuracy of 97.3%, followed by Information Gain with 5 features and an accuracy of 97.3%, and Pearson Correlation with 3 features and an accuracy of 88.0%. The results of this study demonstrate that selecting appropriate features can significantly enhance an ANN's ability to categorize sleep problems. The proposed approach is expected to be a reference in the development of a more accurate sleep disorder diagnostic aid system.]]>
Copyrights © 2026
