Credit_Risk_Analysis

Overview of Analysis

For this analysis, I need to solve a credit card risk challenge using machine learning. Since credit risk is an unbalanced classification problem, I need to employ different techniques to train and evaluate models with unblanaced classes. I will use the imbalanced-learn and scikit-learn libraries to build models using resampling. Then, I will oversample the data using RandomOverSampler and SMOTE, and undersample the data using ClusterCentroids. I will then combine the over and undersampling methods using SMOTEENN. Lastly, I will compare two machine learning models, BalancedRandomForestClassifies and Easy EnsembleClassifier, to predict credit risk.

Purpose

The purpose of this analysis is to determine whether these machine learning models should be used to predict credit risk.

Results

RandomOverSampler Model

• The balanced accuracy score is about 65%.

• The high_risk precision is 1% and its recall score is 63%.

• The low_risk precision score is 100% with a recall of 67%. The precision score is very high because the low_risk population is high.

SMOTE Model

• The balanced accuracy score is 61%.

• The high_risk precision is 1% and the recall is 55%.

• The low_risk precision is 100% and the recall is 66%. The low_risk population is high, so the precision score is high.

ClusterCentroids Model

• The balanced accuracy score is 52%.

• The high_risk precision is 1% and the recall is 60%.

• The low_risk precision is 100% and the recall is 45%. The recall is only 45% because of a high number of false positives.

SMOTEENN Model

• The balanced accuracy score is 65%.

• The high_risk precision is 1% and the recall is 75%.

• The low_risk precision is 100% and the recall is 56%. The recall is only 56% because of a high number of false positives.

BalancedRandomForestClassifier Model

• The balanced accuracy score is 79%.

• The high_risk precision is 4% and the recall is 67%.

• The low_risk precision is 100% and the recall is 91%. The recall is 91% because of a low number of false positives.

EasyEnsembleClassifier Model

• The balanced accuracy score is 93%.

• The high_risk precision is 7% and the recall is 91%.

• The low_risk precision is 100% and the recall is 94%. The recall is 94% because of a low number of false positives.

Summary

I ran six different tests to determine balanced accuracy, precision, and recall. None of the tests showed high precision in determining if a credit risk is high or not. The high_risk precision didn't get above 7%, which is very low. The best model we used was the EasyEnsembleClassifier model because it has a recall of 91% for high_risk, but the precision was still low at 7%. Since the precision is low, a lot of low_risk credits are falsely detected as high_risk, which is not good. Overall, none of these models are good enough to predict credit risk.