Predictors of a Depression Indicator in a Large Public Dataset: Logistic Regression and Neural Network Comparison

Depression ranks among the five leading disability causes. The research associations with depression include many demographic, lifestyle and health-related measures. These factors and depression probably interact with each other.

The objectives of the study are to describe the characteristics of the participants, to examine the bivariate association between a proxy of depression and the study variables, to identify independent predictors with the help of a logistic regression model, and to evaluate a neural network classification model.

A cross-sectional secondary analysis of a publicly available dataset N = 50000 was conducted. Depression indicator (Yes/No) is outcome variable. A neural network of the multilayer perceptron type (15 inputs, a hidden layer with.

Depression indicator in 40.7% of participants. In bivariable analyses, physical activity, alcohol consumption, dietary habits, family history of depression, and chronic medical conditions showed statistically significant but small differences. Income (p = 0.239) and number of children (p = 0.078) did not differ between groups.

There was a modest association with depression index and poor discrimination by neural networks. In performing a classification relating to depression using machine learning one must use assessments with validated outcomes and employ class-sensitive performance assessment.