Steve Filippelli: Random Forests

Random Forest is a versatile model that is quick and easy to implement, robust to overfitting, and produces reasonable accuracy with minimal tuning. We'll look at the basics of how this model works and some of the beneficial attributes which make it useful for classification and regression problems as well as variable selection. Then we'll walk through an application of classification by mapping land cover type from satellite imagery, and (time willing) we'll also go through a regression example by mapping percent tree cover. Through these examples we'll see the influence of hyperparameters on model accuracy and processing time, the effect of multi-collinearity on variable importance, and tools for drawing inference such as partial dependence plots.