Prediction of Sleep Disorder: Insomnia Using AdaBoost Ensemble Learning Algorithm with Grid Search Optimization

Abstract

Human health is an important thing to keep. Health has to be maintained with appropriate rest. Lack of rest has a bad impact on the body such as hormonal imbalances. One of the causes of lack of rest is insomnia. Insomnia is a phenomenon that describes someone's difficulty sleeping. Insomnia is often considered trivial, but chronic insomnia puts the sufferer at risk of serious illness physically and psychologically. Some people sometimes don't realize that they have insomnia because they feel like they have trouble sleeping. Therefore, early detection of insomnia is necessary to do. This study uses a machine learning approach to make predictions, namely the AdaBoost + grid search method. AdaBoost is used because of its reliability in making strong classifiers and grid search is applied to tuning parameters from AdaBoost. Parameters that are optimized are the n estimator and learning rate. Parameter tuning by grid search gives n – estimator = 76 and learning rate = 0.1. Some preprocess technique is done, there are normalization and ordinal encoding then data splitting based on the determined ratio. There are 80% for training data and 20% for testing data. On training data, the result is 98% percentage for each accuracy, precision, recall, and f1 score. This value is better than the comparison method, it is LogRegression that only reaches 97% value on each evaluation measure. The model implemented on test data and AdaBoost + grid search obtained 100% accuracy, precision, recall, and f1 score. However, LogRegression only gives 98% result. This study proved that AdaBoost with grid search is sustainable to do early prediction of insomnia.