K-Nearest Neighbors (page 5 of 6)

We evaluate the quality of a KNN classification model using the confusion matrix and different error metrics for regression problems. For binary classification problems, we may also use the ROC curve. Remember, we want to balance bias and variance when selecting the best KNN model with the optimal 'k' and distance hyper-parameters. Often, a larger 'k' might lead to overfitting, but a smaller 'k' might lead to under-fitting.