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

Area of Science:

• Public Health
• Computational Psychiatry
• Epidemiology

Background:

• Depression is a leading cause of disability globally.
• Numerous demographic, lifestyle, and health factors are associated with depression.
• These factors likely interact with depression, necessitating comprehensive analysis.

Purpose of the Study:

• Characterize participants in a large dataset.
• Examine bivariate associations between depression indicators and study variables.
• Identify independent predictors of depression using logistic regression and evaluate neural network models.

Main Methods:

• Cross-sectional secondary analysis of a dataset with 50,000 participants.
• Depression status (Yes/No) as the primary outcome variable.
• Multilayer perceptron neural network model employed for classification.

Main Results:

• A depression indicator was present in 40.7% of participants.
• Physical activity, alcohol consumption, diet, family history, and chronic conditions showed significant associations.
• Income and number of children did not show statistically significant differences.

Conclusions:

• Neural networks demonstrated modest association and poor discrimination for depression prediction.
• Accurate depression classification using machine learning requires validated outcome assessments and class-sensitive performance evaluation.