Error rate, also called misclassification error or classification error, measures the proportion of classification predictions that are wrong. It adds false positives and false negatives, then divides by the number of records: (FP + FN) / n. Model accuracy is its complement: 1 − error rate. section 14.6.2
Calculate and interpret the metric from binary predictions:
- Create a confusion matrix by comparing actual outcomes with predicted classes. section 14.6.2
- Count false positives and false negatives, add them, and divide by the test-set record count. section 14.6.2
- Compare the result with model accuracy, while considering whether false positives or false negatives matter more in context. section 14.6.2
NBA postseason prediction
For a logistic regression predicting NBA playoff qualification, the confusion matrix contains 11 false positives and 10 false negatives among 80 test records. In R, `misclassification_error <- (11 + 10) / 80` returns 0.2625, or a 26.25% error rate. The corresponding accuracy is `1 - misclassification_error`, which returns 0.7375, or 73.75%. section 14.6.2
confusion_matrix <- table(actuals$actual_values, predictions$predicted_values2) print(confusion_matrix) ## 0 1 ## 0 19 11 ## 1 10 40
Inspect the confusion matrix behind the NBA example, including the 11 false positives and 10 false negatives used in the error-rate calculation. section 14.6.2
