A rule induction based credit appraisal system
Kenyan based institutions are facing a major problem today as they are handling a lot of data which they have collected over the decades. Kenya has in the past decade experienced growth in the industry sector. Most companies have piled up customer data especially in service-based industries. It is not possible for the human mind to be able to process this information and be able to make real-time decisions based on realtime reports. Often, the precious data is not even utilized to assist in decision making due to lack of the knowhow on handling the data. Data mining is the process of extracting patterns from data and is becoming an increasingly important tool to transform this data into information. This research aimed at exploring the various algorithms used in the data mining tools and use rule induction to build a model that can be used to appraise customers for a credit facility. The rules are built from data based on statistical significance calculated from the analysis of variable selection and significance. Weightings on each variable are obtained by use of statistical methods (Regression, Information value and Weights of evidence) and a lot of expertise advice. The model has the ability to alter cell values in real time, in line with what the end user input or does on the dashboard. To be able to do a gauge so that the model gives a good or bad rating, a risk meter was designed and a minimum value was given. The risk meter uses the total weight output by the model after accepting a user's inputto give a rating of bad or good. The model was found to have an average accuracy level of close to 60% in its predictions of a good and bad credit borrower. It also gave insight into the data held that goes along way into helping the organization make decisions on areas that are business problems. For example, it was found that male borrowers have a higher defaulting rate than female have, yet less than twenty percent of borrowers were female. For the bank to be able to predict the probable behavior of a customer's loan performance, they must have the history of the client's financial data. This may need an extension to borrow from the introduced Kenyan Banks Credit Bureau database so as to have a three-sixty degree view of a customer.