Probabilistic Scoring Model for Identifying Bank Clients

This diploma thesis focuses on developing a probabilistic model for quantitatively evaluating the quality of client data in a banking environment, specifically targeting Master Data Management. The goal was to create a model that enhances the data matching process by calculating a score that identifies data uniqueness within a range of 0 to 100. This model is intended as a supplementary system to the existing unification rules that govern data matching. The thesis begins with a general overview of Master Data Management and its implementation within the bank, emphasizing client data matching. The development of the model followed the CRISP-DM methodology, including the design of the calculation concept, selection of relevant attributes, generation of frequency profiles, and creation of a dual weighting system. The first weighting system was based on clustering the frequencies of individual attributes, while the second was determined through optimization tasks that considered statistics and expert opinions. The resulting score, formed by summing the multiplied weights of all attributes, was applied to test data and analyzed, including the establishment of a threshold value for clearly separating unique records. Validation of the model through score development analysis based on various parameters and expert interviews confirmed its functionality and potential for improving client data quality and process efficiency. The thesis proposes a web service to enable automation of calculations and future integration of the model into banking systems, contributing to more efficient decision-making during client onboarding, reducing duplication risk, and enhancing the reliability of data management in the bank.