BASED ON BACKFLOW LEARNING, A NEW NOISE-ADAPTIVE TWO-LAYER ENSEMBLE MODEL FOR CREDIT SCORING

Abstract

It has become more crucial to use machine learning and artificial intelligence algorithms in categorization challenges like credit scoring. Commercial banks and loan lenders have an essential information management task: building an ensemble learning model that has been demonstrated to be more accurate and resilient than individual classifiers. Extreme gradient boosting (EGB), gradient boosting decision tree (GBT), support vector machine (SVM), random forest (RF), and linear discriminant analysis are all combined into an unique noise-adapted two-layer ensemble model for credit scoring based on backflow learning. Base classifiers may be boosted in strength and variety by using a novel backflow learning technique that retrains them on misclassified data points. Two-layer ensemble modelling is used to combine the predictions of all of the basic classifiers to provide a final predictive result. As a result, an isolation forest-based noise adaptation strategy is being suggested to solve the issue of noise data, which has long been considered a key impediment to the accuracy of prediction models. Outlier scores are calculated for each data point in the training set to identify noise data, which are then boosted to generate the noise- adapted training set. A comparison of the proposed model's performance against that of existing benchmark models is carried out using three credit datasets from the UCI machine learning library. Our suggested model outperforms other models in terms of several performance metrics, according to the findings of our experiments.