Random Forest Model

Random Forest is a highly configurable model. We started our tuning exercise with 200 trees, which we were able to reduce to 10 while still maintaining a good accuracy. Next we tried setting the max_features to ‘sqrt’ to limit the number of features being considered for each split. Given the wide variety of parameters, we went with a Grid Search. It is very resource consuming and takes about 1min runtime with an M1 chip, however does not improve the accuracy drastically and so was not considered as a candidate for our predictor app.

Random Forest Model with Full 30 Features