Lydia Delgado Uriarte
Evaluate machine learning models or algorithms to predict credit risk.
-
Balanced accuracy score: 0.7877 -> 79%
-
Sensitivity/recall: 0.67 A low recall is indicative of a large number of false negatives.
-
Out of 87 actual high risk , 58 were predicted to be high risked, which we call true positives.
-
Out of 87 actual high risk, 29 were predicted to be low risk, which are considered false negatives.
-
Balanced accuracy score: 0.9254 -> 93
-
Sensitivity/recall : 0.91 Highest recall of all, meaning high prediction can be likely true negatives.
-
Out of 87 Actual High risk 79 were predicted to be high risked, which we call true positives.
-
Out of 87 Actual High risk, 8 were predicted to be low risk, which are considered false negatives.
-
Balanced accuracy score: 0.6533 -> 65%
-
Sensitivity/recall : 0.61 A low recall is indicative of a large number of false negatives.
-
Out of 87 actual high risk , 53 were predicted to be high risked, which we call true positives.
-
Out of 87 actual high risk, 34 were predicted to be low risk, which are considered false negatives.
-
Balanced accuracy score: 0.6512 -> 65%
-
Sensitivity/recall : 0.62 A low recall is indicative of a large number of false negatives.
-
Out of 87 actual high risk , 54 were predicted to be high risked, which we call true positives.
-
Out of 87 actual high risk, 33 were predicted to be low risk, which are considered false negatives.
-
Balanced accuracy score: 0.5103 -> 51 %
-
Sensitivity/recall : 0.64 A low recall is indicative of a large number of false negatives.
-
Out of 87 actual high risk , 56 were predicted to be high risked, which we call true positives.
-
Out of 87 actual high risk, 31 were predicted to be low risk, which are considered false negatives.
-
Balanced accuracy score: 0.6375 -> 64 %
-
Sensitivity/recall : 0.70
-
Out of 87 actual high risk , 61 were predicted to be high risked, which we call true positives.
-
Out of 87 actual high risk, 26 were predicted to be low risk, which are considered false negatives.
The majority of the models has between the range of 50% - 80% accuracy and the sensivity too. In this case is to take in consideration the sensitivity of each models.
The best model is the Easy Ensemble AdaBoost Classifier due the sensivity and acccuracy is also important to predictions. Itβs more important to detect potentially fraudulent transactions, high sensitivity means that among people who actually have credit risk, most of them will be correct and the problem would be treated right away.