HarvardX PH125.9x: Higgs dataset classification

In this project, neural networks (via Keras in R) was used to predict particle collision events using the HIGGS dataset. The final NN was generated with three hidden layers of 2048 nodes each, generating a final area under the ROC curve (AUC) of 0.877. It was found that using high-level (derived) features provided only a small improvement in AUC, compared to using low-level features only.