Enhancing Sleep Disorder Prediction: CNN with IQR and Lasso Preprocessing

Millions of people worldwide suffer from sleep disorders, which spotlight the significance of precise diagnosis and prognosis. Sleep disorder problem is nowadays a major problem in adults which make them unsleep thus leading to many complications. The main scope of this research is to predict sleep disorder in a well efficient manner using machine learning approaches. To improve the predictive accuracy for the classification of sleep disorder, we present a unique method in this study that combines the application of Lasso feature extraction and Convolutional Neural Network (CNN) with Interquartile Range (IQR) preprocessing. The data quality was enhanced and the impact of anomalies on model performance was reduced by preprocessing the dataset first. The preprocessing procedure was done using IQR, a reliable technique for the identification and removal of anomalies. At the same time, Lasso feature extraction is implemented to choose the more related attributes, improving interpretability, and minimizing dimensionality. The CNN algorithm is well-known for its capability to spontaneously read hierarchical attributes derived from the dataset. The model is accelerated with the preprocessed attributes extracted from the Lasso and IQR combination. The incorporation permits CNN to acquire both global and local designs in the data, increasing its ability to discriminate between different types of sleep disorders. The algorithm is trained with the help of the ISRUC-Sleep dataset. The outcome of the algorithm is estimated concerning precision, accuracy, F1-score, and recall scales. The results obtained from the experiment illustrate that the suggested combination of Lasso and IQR preprocessing along with CNN performs well than any other individual algorithms based on exactness of 90.30%, precision value of 0.89, recall rate of 0.87 and F1-score of 0.90 respectively. The software tool used is Jupyter Notebook and the Python programming language is used.