Abstract
The credit card has been one of the most successful and prevalent financial services being widely used across the globe. However, with the upsurge in credit card holders, banks are facing a challenge from equally increasing payment default cases causing substantial financial damage. This necessitates the importance of sound and effective credit risk management in the banking and financial services industry. Machine learning models are being employed by the industry at a large scale to effectively manage this credit risk. This chapter presents the application of the various machine learning methods like time series models and deep learning models experimented in predicting the credit card payment defaults along with identification of the significant features and the most effective evaluation criteria. This chapter also discusses the challenges and future considerations in predicting credit card payment defaults. The importance of factoring in a cost function to associate with misclassification by the models is also given.
TopIntroduction
Credit cards are merely plastics issued by banks and financial institutions. Credit cardholders can use these to make retail purchases physically in stores or digitally from the convenience of their home. Credit card transactions are getting more secured by the day and gradually replacing cash transactions. These credit cards are part of the credit lending business that has been highly profitable for the banking and financial services industry. However, this surge of demand comes with the increased risk of payment defaults.
A payment default is a condition where the credit card holder does not pay back the amount owed to the bank within the payment due date. Payment defaulters impose a significant loss to the lenders globally by not paying back the owed amount and forcing banks to write off these amounts as non-recoverable bad debts in extreme cases. This deep worrying concern necessitated the establishment of a sound risk management system that can predict defaulters in advance and alert banks enabling them to take corrective action thereby mitigating risk. Machine learning models are being employed by the industry at a large scale to effectively manage this credit risk. These machine learning models can identify the significant features and effectively predict the payment defaults in advance.
The existing studies have identified gender, age, marriage, education, payment related variables, repayment related variables, employment status, and limit balance as the most significant features that impact the default prediction (Choubey, 2018; Leong & Jayabalan, 2019; Neema & Soibam, 2017; Sariannidis et al., 2020; Sayjadah et al., 2018; Ullah et al., 2018; Xu et al., 2017; Xu et al., 2018). Leow & Crook (2014) derived features like relationship duration with a bank, transitions between the states of delinquency, average payment amount, and average repayment amount to increase the model performance. Further, few studies have considered macro-economic variables that could potentially impact defaults prediction such as bank interest rate, Consumer Price Index (CPI), Gross Domestic Product (GDP), and unemployment rate (Bellotti & Crook, 2013; Li et al., 2019).
Numerous researchers have shown the application of time series, machine learning, and deep learning models to increase the accuracy of default prediction and minimize loss. The time-series models were implemented on the data spanning across few years to capture the economic conditions. Further, these datasets had several features that could be further classified into static and dynamic features. The factors contributed to the superior performance of time-series models with accuracy close to 95% along with high precision and recall rates (Bellotti & Crook, 2013; Ho Ha & Krishnan, 2012; Leow & Crook, 2014). Machine learning models are typically preferred over time-series models when datasets are not very huge and the period of data does not span over a longer period. Machine learning models like Logistic Regression (LR) have been used for simplicity of implementation and interpretation. Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Artificial Neural Network (ANN), and Recurrent Neural network (RNN) were able to achieve accuracy close to 90% with an equally satisfying precision and recall values (Leong & Jayabalan, 2019; Mazumder, 2020; Neema & Soibam, 2017; Sariannidis et al., 2020; Sayjadah et al., 2018; Ullah et al., 2018; Yontar et al., 2020).
Few studies have experimented with deep learning models as well and achieved high model accuracy (Chishti and Awan, 2019; Ebiaredoh-mienye et al., 2020; Hsu et al., 2019; Sun and Vasarhelyi, 2018). These deep learning models are artificial neural networks that have more than one hidden layer to transform the non-linear relationships effectively. An artificial neural network is an attempt to mimic the functioning of a human brain. Details of these studies are elaborated in a separate subsection.
Key Terms in this Chapter
Hyperparameter Tuning: This is a process of fine-tuning the hyperparameters to achieve optimal model performance. This process typically involves randomized-search or grid-search methods applied in 5-fold or 10-fold cross-validation regions.
Misclassification Cost: Cost associated with the wrong classification of an event. In this chapter, the event is “payment default”. Hence, this misclassification cost is assigned to the model that wrongly predicts payment default as a non-default (Type-2 error which is more dangerous) and non-default as a payment default (Type-1 error).
Significant Features: These are the independent or explanatory features or variables that can significantly explain the dependent feature or variable.
Payment Default: Inability of the customer to make payment of the total due amount owed to the issuer by the specified due date. This applies to credit cards, housing loans, education loans, personal loans, or any similar product offered by banks or financial institutions.
ADASYN (Adaptive Synthetic): An algorithm that generates synthetic data, and its greatest advantages are not copying the same minority data, and generating more data for “harder to learn” examples.
Cross-Validation: This is a region formed out of the Training dataset with the primary objective of hyperparameter tuning the machine learning models to achieve the best performance. This entire process of hyperparameter tuning using a cross-validation region is called a cross-validation process.
Class Imbalance: When the observations or data points associated with an event is rare as compared to the data points or observations associated with the non-event, then the situation is called an imbalanced class. Typically, the data distribution percentage of data points in such cases range from 99:1 to 70:30 between majority and minority class respectively. The data which is available for typical classification problems like fraud detection and default prediction have such severe class imbalance.
SMOTE (The Synthetic Minority Over-Sampling Technique): An oversampling approach that creates synthetic minority class samples.
Balancing Techniques: These are techniques employed to overcome the class imbalance problem in a given dataset that significantly impacts the performance of the machine learning model. The techniques either employ over-sampling or under-sampling methods to overcome the balancing problem. Some of the popular balancing techniques are ROS, SMOTE, and ADASYN.
Hyperparameter: These are model parameters that can be manually configured (through expert judgment or empirical results) to achieve optimal performance. Every machine learning model has its own set of hyperparameters.
Credit Card: Plastic cards issued by Banks or Financial institutions with a pre-assigned credit limit for customers to make purchases at retail stores (also called merchants) or even online.
RIPPER (Repeated Incremental Pruning to Produce Error Reduction): Algorithm introduced by W. Cohen in 1995, which improved upon IREP (Incremental Reduced Error Pruning Algorithm) of Furnkranz and Widmer in 1994 by generating rules that match or exceed the performance of decision trees. Having evolved from several iterations of the rule learning algorithm, the RIPPER algorithm can be understood in a three-step process: Grow, Prune, Optimize.