HOPS: Hyperparameter optimization and predictor selection

We developed the hyperparameter optimization and predictor selection (HOPS) software to optimize hyperparameters and predictor selection while limiting correlation among the selected predictors for machine learning models. Including correlated predictors in machine learning models can distort model estimation and prediction and introduce bias in predictor importance estimates. The HOPS software excludes unnecessary predictors that do not improve model performance and limits noise introduced when including correlated predictors. It can improve model performance while reducing the data required for predictions and decreasing processing time. The original idea for this software is the approach used to fit gradient-boosted decision trees for the Landsat Burned Area algorithm. We designed HOPS to work with regression and classification models in the scikit-learn machine-learning library.