A novel machine learning-based prediction method for patients at risk of developing depressive symptoms using a small...

Area of Science:

• Mental Health Research
• Machine Learning Applications
• Biomarker Discovery

Background:

• Depression prediction is a priority for early intervention and treatment success.
• Self-reported feelings offer a valuable, low-dimensional network biomarker for depression.
• Network data provides a compact representation of high-dimensional information for machine learning.

Purpose of the Study:

• To predict depression-prone patients using network-formatted self-reported logs.
• To evaluate the effectiveness of the graph convolutional network (GCN) algorithm for this prediction task.
• To address challenges with small datasets in biomarker research.

Main Methods:

• Applied the graph convolutional network (GCN) algorithm to network-formatted self-reported log data.
• Utilized a data augmentation technique to expand a small initial dataset.
• Tested the model across three experimental cases with varying ratios of depressive cases.

Main Results:

• Achieved high prediction accuracy ranging from 86-97%.
• Obtained F1 scores between 0.83-0.94 across all experimental cases.
• Demonstrated consistent high performance regardless of the proportion of depressive cases.

Conclusions:

• Self-reported logs combined with GCN show significant potential for early depression prediction.
• The approach is effective even with limited data, a critical factor in biomarker research.
• This method offers a promising avenue for advancing depression prediction and warrants further investigation.