MAPS: Mobile Assessment for the Prediction of Suicide
- Goal: Prediction of Suicidal Thoughts and Behaviors in Youth Using Intensive mobile sensing
- Hypothesis Driven Data analysis
- Interpretable features (e.g., Sleep duration, Sentiment in language, Music choice, Frequency of social contact)
- Dynamic factor analysis to detect underlying temporal structure of time series data
- Multilevel regression techniques (i.e., autocorrelation/cross-correlation models)
- Mixed Markov models for predicting state-switching (i.e., changes in suicide risk occurring within a person over time)
- Empirically-Driven Data Analysis (Machine Learning)
- Will explore all available data (hypothesized and non-hypothesized) using deep learning models
- Computational feature representation to extract features understandable by the computer from the raw input streams
- Concurrent fusion to represent co-occurring visual, acoustic, verbal, and behavioral features (i.e., to model synchronized behaviors) and
- Temporal fusion to model the temporal patterns and contingencies between the multimodal behaviors (i.e., to model asynchronized behaviors such as changes in suicidal risk)
Project Person
