Enhancing Classifier Performance Using a Spatial Perspective Based Ensemble Model: An Application to Credit Scoring

Abstract

Credit scoring can be defined as an applied mathematics procedure that forms the basis for several financial establishments to make the most appropriate decision concerning the extension of monetary benefits to customers based on careful analysis of their transaction history with the intent of cutting down the organization’s expenditure and eliminate potential risks. Immaterial attributes typically degrade the classification accuracy, so relevant feature selection will facilitate to deal effectively with large datasets. Heterogeneous ensemble-based models have outshone other statistical and Artificial intelligence-based techniques proposed for the problem hence we have used heterogeneous ensemble for credit scoring preceded by a novel algorithm for feature selection. An evolutionary algorithm called Binary cuckoo search has been used for the extraction of the most significant features from the dataset. Once the most optimal set of features is obtained, classification is done using four different base classifiers. The results predicted by each of these unique base classifiers are then combined using a novel methodology of ensemble modelling and weight determination. During performance evaluation, our approach has shown appreciable results on both the benchmarked credit scoring datasets- Australian and German dataset as compared to other ensemble-based methods.