Developing Intelligent Systems for Credit Scoring Using Machine Learning Techniques
PhD thesis by: Bart Baesens (2003), KATHOLIEKE UNIVERSITEIT LEUVEN
The last decades witnessed a spectacular increase in computer infrastructures and resources to store huge amounts of data. Many businesses have eagerly adopted these data storing facilities to record information regarding their daily operations. Examples are banks that store information regarding the repayment behavior of their customers, supermarkets that store every purchase of an individual into their data warehouses and stock exchanges that record stock prices at regular time inter- vals. These are all businesses where massive amounts of data are being generated and stored electronically on a daily basis.
Until recently, this data was analyzed using basic query and reporting utilities. The advent of knowledge discovery in data (KDD) technology has created the opportunity to extract more intelligent and advanced knowledge from these huge collections of data. Machine learning is one of the key technologies underlying KDD and aims at acquiring knowledge by learning patterns from data. Knowledge of these patterns could then be efficiently used to optimize sales strategies or business operations in order to gain profits or cut costs.
In this dissertation, we study the use of machine learning algorithms to develop intelligent systems for credit scoring. The problem of credit scoring is essentially a classification task which aims at distinguishing good payers from bad payers us- ing all possible characteristics describing the applicant. It is our aim to develop decision support systems for credit scoring which are both accurate and comprehen- sible. Both criteria are believed to play a pivotal role in the successful deployment of automated credit scoring systems.