Identification of depression predictors from standard health surveys using machine learning

Abstract

Depression has profound personal, societal, and economic impacts. Leveraging advances in technology can help identify predictors of depression. In this study, we compared seven machine learning (ML) algorithms to identify depression predictors using publicly available datasets from standard health surveys. We obtained data from the National Health and Nutrition Examination Survey (NHANES) 2017–2020, including medical, mental, demographic, and lifestyle information from 8965 individuals aged 18 to 80 years. Our study identified strongly correlated features of depression and demonstrated that ML algorithms can accurately identify depression predictors. The performance of the algorithms was evaluated using standard metrics. Among the algorithms tested, the Neural Network algorithm showed the highest overall performance, with an area under the curve of 91.34 %, which significantly outperformed results obtained with traditional statistical methods such as logistic regression and nomograms. This study demonstrates how ML can mine standard health surveys and identify depression predictors in a more accurate and nuanced fashion than other approaches. The findings of this study further suggest that incorporating heterogeneous data can enhance the performance of ML algorithms. These algorithms can be a valuable complementary tool for clinicians, particularly in remote settings, facilitating data analysis, and accelerating knowledge discovery in mental health studies.

Keywords

• Depression
• Depression predictor
• Machine learning (ML)
• NHANES
• Feature Selection