ENV330H5 Lecture Notes - Lecture 6: Long Term Ecological Research Network, Anvers Island, Palmer Station
Document Summary
Important thing is not how well the model fits the training data it"s how well it works on the testing data. How well does each variable do at creating a homogeneous data set (i. e. , dividing people into who does & doesn"t like psls?) 2 candidates: likes autumn likes sweets & like starbucks . Calculate the impurity of each leaf arising from a root node, and take the weighted average of the impurity of the leaves (weighted by the number of observations in each leaf). The impurity of a leaf = 1 (probability of yes )2 (probability of no )2. The lower the gini number, the purer the node. This is the overall correctness of the model the proportion of correct answers. Sensitivity= proportion of positives that were correctly identified. = true positives / (true positives + false negatives). Remember, a false negative is a positive that was misclassified. = true negatives/ (true negatives + false positives)
