A Variable Neighborhood Search Combined with EM Clustering and Classification: Application to Credit Assessment

Data mining involves a range of techniques designed to improve logistics and supply chain management. It has great significance and importance in logistics enterprises. It can be used in production, transportation, supplier selection and warehousing. Also, it can optimize the supplier selection, facility location, vehicle routing and help managers in decision making. Clustering, or partitioning, is a prominent task in data mining aimed at dividing large datasets into homogeneous groups, where items within the same group exhibit similarity. This work focuses on credit scoring to assist banks and managers in their decision-making processes. We use Expectation-Maximization (EM) clustering to categorize applicants, and subsequently apply a k-NN classifier to assess each applicant's creditworthiness based on the clustered data. To boost the effectiveness of our method, we introduce a feature selection strategy. We suggest employing a variable neighborhood search (VNS) technique to identify the most relevant features for the clustering task. We evaluate our methods using several financial datasets, and the results demonstrate their effectiveness. Notably, our findings indicate that VNS significantly improves performance, resulting in increased accuracy across all analyzed datasets.