A hybrid interpretable credit card users default prediction model based on RIPPER

Abstract

Summary With the vigorous development of the financial sector, financial risks are showing a tendency toward diversification, particularly regarding the customer credit risk of commercial banks. Therefore, the customer's credit risk is being considered by financial institutions, and a credit evaluating model has emerged as a result. Currently, research has concentrated on enhancing the precision of the model, ignoring the interpretability, which makes it difficult to apply in the industry. Compared to precision, studies related to the interpretable model are limited. In our previous work, we did not consider model operation time and stability. Therefore, this study proposes a hybrid model based on the RIPPER algorithm. First, according to the characteristics of credit card data sets, targeted special data pretreatment methods are proposed. Next, the RELIEF method for feature selection removes the redundant features and further improves the interpretability of the model. Then, to address the problem of the imbalanced distribution of credit card data sets, a synthetic minority class sampling algorithm is used to equalize the samples. Finally, default credit card users are predicted by taking advantage of the rules generated by the RIPPER algorithm. To test the performance of the model, we used Taiwanese credit card customer data for empirical research. We considered model accuracy and interpretability when comparing the proposed SPR‐RIPPER model with the existing mainstream models. The results of the experiments indicate that the proposed model achieves acceptable results. This study demonstrates that the proposed credit card user default prediction model, SPR‐RIPPER, has practical application value.

An Investigation of Credit Card Default Prediction in the Imbalanced Datasets

“…It has also been observed that in previous studies machine learning models, i.e., random forest [11], stacking [16], the GBDT model [22], logistic model tree [23], bagging [24], and k-nearest neighbor [26] were not given efficient results on imbalanced data on credit card default prediction data, as shown in Figure 5. In previous research, Random forest [11] was given the accuracy of 58.8%, but while combining the random forest with K-means SMOTE oversampling technique, the result has been significantly improved with the accuracy of 88.2%.…”

“…Various imbalanced datasets like lending club dataset [13,14], Chinese P2P lending company dataset [15], German credit dataset, Australian credit dataset, and Dataset of We.com [16], Chinese consumer finance company dataset [17] were used in the past. Previous studies [2,11,13,[15][16][17][18][22][23][24][25][26] were not deployed in the models for end-users.…”

“…To the best of our knowledge, there was no work done on the default credit card client's dataset by employing various resampling techniques. Several studies reported that effective results were not obtained [11,[22][23][24][25][26] when imbalanced data has utilized. Contributions: This research possesses various contributions in the domain of credit risk prediction.…”

High performance deep learning techniques for big data analytics

“…A portion of the papers included in this special issue focus on traditional machine learning. Xu et al present a hybrid interpretable model for fraud detection in user credit transactions. Yan et al optimize a neural network with genetic algorithms for finance early warning in the insurance sector.…”

“…A portion of the papers included in this special issue focus on traditional machine learning. Xu et al 2 It features online transfer learning by utilizing the extracted knowledge from the current state of the tracking targets for generating tracking decisions. Petri Nets are also used for supervised and unsupervised learning.…”