A machine learning approach predicts future risk to suicidal ideation from social media data

This study developed an algorithm, SAIPH, to predict future suicidal ideation risk using publicly available Twitter data. Neural networks were trained on Twitter data related to psychological constructs (burden, stress, loneliness, hopelessness, insomnia, depression, anxiety). A random forest model, using neural network outputs, predicted suicidal ideation status with an AUC of 0.88. The model also predicted temporal risk, showing a seven-fold increased risk within 10 days of a peak in individual-specific risk scores. Validation using regional Twitter data showed significant associations between algorithm scores and county-wide suicide death rates, particularly in younger individuals. SAIPH shows potential as a clinical decision tool for suicide risk screening and monitoring.

Arunima Roy, Katerina Nikolitch, Rachel McGinn, Safiya Jinah, William Klement, Zachary A. Kaminsky