measuring models

Recall: true positives / true positives + false negatives - 'true positive rate'

 

Precision: true positives / true positives + false positives - 'correct positives'

 

Specificity = TN / (TN+FP) - 'true negative rate'

 

F1 score = 2TP / (2TP + FP + FN)

 

RMSE (Root mean square error) - accuracy measurement

 

ROC curve (Receiver Operating Characteristic Curve) - 좌상향으로 더 굽어질수록 좋은 모델

 

AUC (Area Under the ROC Curve)

 

P-R Curve (Precision / Recall Curve) - 곡선 아래 넓이가 넓을수록 좋은 모델