Logistic Regression Lab

Goal

In this lab, we’ll again, try to predict players’ skill level, i.e., the SkillLevel column, which takes integer values between 0 and 3, with 0 being novice and 3 being expert. As we have more than two values for the class variable SkillLevel, we’ll apply multinomial logistic regression, that is, logistic regression for more than two class values, instead of the vanilla version that only deals with binary classification cases. We’ll use the logistic regression implementation in the nnet package, since the one in the caret package isn’t as robust.

Brief refresher

Similar to Naive Bayes, Logistic regression estimates class probabilities, instead of directly predicting the class like KNN. Logistic regression assumes that the probabilities can be approximated by passing a linear combination of independent variables’ values through a logistic function. You can therefore see where the linear assumption and the logistic part of its name comes from.

Preparation

Though we will not use the caret package for the logistic regression model, we'll still import it to leverage the preprocessing capabilities.