garyforeman / magic-gamma Goto Github PK

Generate a simple model as a reference point. Use either knn or logistic regression and compare to results given in Bock et al. (2004).

Gradient boosted machines or an artificial neural network would make good candidates for this model. Should be able to beat the AUC of the baseline model.

Deploy the production model using AWS and create an API that will accept features and report back a prediction as to whether the event was a photon or hadron.

Explore distributions of feature variables and their correlations with the target. Is there collinearity that should be removed? Is there missing data or outliers?