Credit_Risk_Analysis

Project Overview

Assist the client to predict credit risk using a variety of Resampling Models and algorithms.

Methodology

Tools/Programs/Languages used:

• imbalanced-learn library
• scikit-learn library
• ETL
• Machine Learning

I oversampled data via RandomOverSample and SMOTE algorithms using credit card credit dataset from LendingClub. ClusterCentroids algorithm was used to undersample the data. SMOTEENN algorithm was used for over-and-undersampling. Then I used two machine learning models to predict credit risk.

Summary of Results

• Accuracy score of Random Oversampling using RandomOVersampler
• Accuracy score was 0.657 which means that the model was correct 65.7% of the time.

• Accuracy score of SMOTE Oversampling using SMOTE
• Accuracy score was 0.662 which means that the model was correct 66.2% of the time.

• Accuracy score of Undersampling using ClusterCentroids
• Accuracy score was 0.544 which means that the model was correct 54.4% of the time.

• Accuracy score of Combination (Over and Under) Sampling using SMOTEENN
• Accuracy score was 0.644 which means that the model was correct 64.4% of the time.

• Accuracy score of ML model using BalancedRandomForestClassifier
• Accuracy score was 0.778 which means that the model was correct 77.8% of the time.

• Accuracy score of ML model using EasyEnsembleClassifier
• Accuracy score was 0.920 which means that the model was correct 91.1% of the time.

Recommendations

• Overall, the models had pretty average accuracy scores. The clear winner was the ML model using EasyEnsembleClassifier.
• The Accuracy score was much higher compared to any of the other models.