In recent years, behavioral sciences and public health studies have seen a shift toward producing large-scale data sets. As these studies become more complex, utilizing appropriate regression-based techniques and building predictive models becomes very challenging. As such, machine learning algorithms have drawn a lot of attention recently and have emerged as powerful and flexible techniques that facilitate selection of predictors that are most associated with the outcomes of interest and improve the quality of predictions.

Using behavioral data, the aims of this project are to apply machine learning strategies to develop predictive models via performing variable selection that can accurately identify potential risk factors and make predictions of future health outcomes or adverse events, and to explore implementation of machine learning variable selection techniques in the presence of incomplete data.