step 1:
Accuracy: the proportion of the correctly classified instances.
If the accuracy is pretty high, could we say that this is a good model? -> No
e.g. 假设test组有100个人,90个人approved,10个人declined,需要对100个人进行test,极端情况将100个人都赋值approved,在此种情况下accuracy是90%,但我们不能说正确率高,因为10个declined全部预测错误。
step 2:
Precision = TP/(TP+FP)
e.g. Out of the individuals whom the model predicted to be declined, how many were classified correctly?
解释:预测结果是positive并且实际结果为真 占 预测结果是positive 的百分比
step 3:
Recall = TP/(TP+FN)
e.g. Out of the individuals who actually be declined, how many did the classifier classify correctly?
Step 4:
F1 score = 2(precision*recall) / (precision + recall)