anand-gith / creditcardpaypredict

Credit Card Payment Prediction Model

CreditCardPayPredict

Credit Card Payment Prediction Model

-Understand the Business Defaulting the payment is a risk for financial institutions. So using previous history and payment trends of the customers we will build a predictive model which can predict the whether the customer will pay or default the next installment.

-Data Understand 24 attributes – Credit limit, age, education, sex, marriage, payment history details 30000 observations Default Next Payment - Yes = 1 and No = 0

-Data Preparation - Identify and clean data and stage data Divided data into test and train Training Data: 21000 Test Data: 9000

-Modelling - We here use for classifying algorithms like Ensemble Methods: Random forest,Bagging,Boosting

Non Ensemble Methods: LDA,Logistic and k-NN

Conclusion:

Ensemble methods are computationally intensive as multiple trees to be built and the results are averaged from different models we can’t really explain how the model derives the value just by looking variable importance we could figure out which feature is more prominent and along with that the complexity parameters for these methods we have to run multiple trails to identify optimal value to tune the model’s complexity parameter.

Non-ensemble methods are simplistic and less complexity parameters mainly depends on the feature selection and building model with features correlated with the response variable for better performance. LDA is simple enough just with all features, mean and covariance calculated form data it could differentiate classes nearly to ensemble methods. If LDA is used with subset of features it would have performed better than ensemble methods.

Its not always wise to choose the ensemble methods over simplistic non-ensemble methods. The methods should be evaluated for performance and simple interpretable models which are easy to tune parameters of the models and easy to scale for the dataset in hand. Non-ensemble methods always provide a go-to option due to its simplistic and interpretable nature.