Features Selection for Credit Risk Prediction Problem

Credit risk assessment has drawn great interests from both researcher studies and financial institutions. In fact, classifying an applicant as defaulter or non-defaulter customer helps banks to make a reasonable decision. The classification of applicants is based on a set of historical information of past loans. Data sets for analysis may include different features, many of which may be irrelevant to the decision making process. Keeping irrelevant features or leaving out relevant ones may be harmful, causing generation of poor quality patterns that may lead to confusion decision. Determining an appropriate set of predictors is an important challenge in credit risk prediction research which guarantees better decision-making. It is the task of searching the smallest subset of features that provide the highest accuracy and comprehensibility. Thus, this study proposes feature selection-based classification model on credit risk assessment. To this end, five algorithms are applied, Speed-constrained Multi-objective PSO (SMPSO), Non-dominated Sorting Algorithm (NSGA-II), Sequential Forward Selection (SFS), Sequential Forward Floating Selection (SFFS), and Random Subset Feature Selection (RSFS). The selected subset is evaluated based on three classifiers K-Nearest Neighbors (KNN), Support Vector Machine (SVM) and Artificial Neural Network (ANN). Our proposed model is validated using three real-world credit datasets. The obtained results confirm the efficiency of SMPSO-KNN model to select the most significant features and provide the highest classification accuracy compared to existing models.