| dc.description.abstract | Sleep health has become more and more crucial to citizens. Various methods have been proposed
to detect sleep disorders from physiological signals. However, the progress of collecting these signals
has some drawbacks, leading to the idea of applying the data from wearable devices to identify whether
the person has any sleep disorders. In this study, a comparison among various algorithms listed as
following: Extreme Gradient Boosting (XGB), Logistic Regression (LR), Decision Tree (DT), and
Random Forest (RF) was conducted to predict sleep disorders based on 12 input variables named Age,
Gender, Occupation, Sleep Duration, Quality of Sleep, Physical Activity Level, Stress Level, BMI
Category, Systolic Blood Pressure (Upper Blood Pressure), Diastolic Blood Pressure (Lower Blood
Pressure), Heart Rate, and Daily Steps, deducing which was the best-match algorithm. Besides, to
investigate which variables contribute more in the results of predicting sleep disorder, we split 12
initial input variables into 4 subsets based on variables’ correlation scores and variables’ important
scores; more specifically, a subsets included 7 highest correlation input variables, a subsets included
3 highest correlation input variables , a subsets included 7 most important input variables, and a subsets
included 3 most important input variables. Then the model performance was evaluated by accuracy
score and F1-score. XGB showed the perfect fit result to this classification task, with accuracy score
of 96.49% and F1-score of 97.82% before being tuned. Tuning XGB raised the accuracy score and
F1-score, of 98.25% and 98.90% respectively. These results show a promising chance for the model
to be integrated into diagnosis systems as well as testing with data from Vietnamese people. | en_US |
