In the case of KNN with K=1, we have a training error rate of 0% because in this case, we have P(Y=j|X=xi)=I(yi=j) which is equal to 1 if yi=j and 0 if not. We do not make any error on the training data within this setting, that explains the 0% training error rate. However, we have an average error rate of 18% wich implies a test error rate of 36% for KNN which is greater than the test error rate for logistic regression of 30%. So, it is better to choose logistic regression because of its lower test error rate.