Suicide is an increasingly dire issue in the United States, especially among youth. The youth suicide rate has been steadily increasing for at least 20 years. The current mental healthcare system in the US has been shown to be ineffective at predicting or preventing suicide with any reliability. In fact, human mental health professionals are only able to identify suicide risk at a rate that is slightly higher than chance. To remedy this, some have begun exploring machine learning solutions to predicting suicide, which has shown initial promise. Furthermore, some have applied a subset of machine learning algorithms known as neural networks to the task, which have proven even more powerful. There is very sparse literature on this subject, however. This research intends to contribute to the literature gap. Additionally, this research intends to improve upon prior results seen in neural network suicide prediction. Researchers are generally able to consistently get results ranging from 0.9-0.95 AUROC. This research will compare several neural network models at the task of predicting suicidal ideation, with a focus on maximizing AUROC.
