Prediction Framework
We now pursue the use of supervised learning to construct classifiers trained to predict depression in our two user classes. To avoid over fitting, we employ principal component analysis (PCA), although we report results for both all dimension-inclusive and dimension-reduced cases. We compare several different parametric and non-parametric binary classifiers to empirically determine the best suitable classification technique. The best performing classifier was found to be a Support Vector Machine classifier with a radial-basis function (RBF) kernel (Duda et al., 2000). For all of our analyses, we use 10-fold cross validation on the set of 476 users, over 100 randomized experimental runs.