We present a novel classifier, the Forest Deep Neural Network (fDNN), and apply it to disease modeling. The fDNN architecture leverages a supervised forest for feature detection, enabling the algorithm to build complex neural networks with a large number of independent features from sparse feature representations while limiting overfitting. We apply the fDNN to model influenza—more specifically, to predict the occurrence of influenza in a cohort of anonymized real-world patients, aided by Natural Language Processing (NLP) and a virtualized Clinical Semantic Network (vCSN). We report accuracy rates in the 90% range, which is extremely high for the problems of disease modeling and applications of NLP.