Mathematical Biology Seminar Harish Bhat, University of Utah, Mathematics Dept. Wednesday, Nov. 28, 2018 3:05pm in LCB 225 Machine Learning Methods for Suicide Attempt Prediction Abstract: Though suicide is a major public health problem in the US, machine learning methods are not commonly used to predict an individual's risk of attempting or committing suicide. In the present work, starting with an anonymized collection of electronic health records for 522,056 unique, California-resident adolescents, we develop machine learning models to predict suicide attempts. We discuss challenges including (1) severely imbalanced classes (the rarity of suicide attempts), (2) the variable length of a patient's history, and (3) the presence of high-dimensional categorical variables (diagnostic codes). We then give customized solutions to these problems, culminating in a deep neural network model with embedding matrices, convolutional layers, and dense layers. We compare our model's performance against other techniques such as random forests, boosting, and logistic regression. For test set observations where we have at least five visits' worth of data on a patient, our best model achieves both sensitivity and specificity exceeding 0.9.