Uncovering the Link between Sleep, Alexithymia, and Overweight Status: A Machine Learning Approach

Abstract

Excessive weight, including obesity and overweight statuses, has been identified as a significant risk factor for a wide range of severe health problems and the severity of psychiatric symptoms associated with various psychiatric diseases. Our primary objective was to predict overweight status using comprehensive data on sleep-related factors and alexithymia. We employed advanced machine learning models, specifically random forest (RF) and neural networks (NN), to identify the most influential predictive factors in this context. The Stockholm Brain Study focuses on functional brain imaging data obtained from two distinct age groups: participants aged 20-30 years and older individuals aged 65-75 years. In order to assess overweight and obesity status, participants' body mass index (BMI) was calculated using the World Health Organization's defined cut-off points for overweight and obesity. Additionally, the study incorporates alexithymia data to further investigate its role in predicting overweight status. The results showed that RF ML model predicted overweight status with an accuracy of 72 % (sensitivity: 100 %, specificity:30%). The area under curve (AUC) value was 95. Furthermore, NN found that overweight status was predicted with the accuracy of 93 % in the train data and the accuracy of 56 % in the test data (sensitivity:80%, specificity:20%). To validate our results, the cross-validation was performed based on algorithms of RF, and NNs, and resulted in an average accuracy score of 0.70 ± 0.12, 0.54±0.11, respectively. The study findings revealed that the prediction of overweight status exhibited remarkable accuracy, surpassing 70 percent, through the utilization of a RF model.