Predicting state level suicide fatalities in the united states with realtime data and machine learning

This study developed and validated a deep learning approach to predict weekly suicide counts at the state level in the US, using real-time online, social media, and health administrative data. A long short-term memory (LSTM) neural network model was built for each of four participating states (Utah, Louisiana, New York, and Colorado). The model accurately estimated state-specific suicide rates, outperforming autoregressive models using only historical death data. This approach offers the potential for more timely estimates of suicide trends, aiding suicide prevention planning and response.

Devashru Patel, Steven A. Sumner, Daniel Bowen, Marissa Zwald, Ellen Yard, Jing Wang, Royal Law, Kristin Holland, Theresa Nguyen, Gary Mower, Yushiuan Chen, Jenna Iberg Johnson, Megan Jespersen, Elizabeth Mytty, Jennifer M. Lee, Michael Bauer, Eric Caine, Munmun De Choudhury