Identifying likely student dropouts using fuzzy inferencing

Fuzzy logic provides a methodology for reasoning using imprecise rules and assertions. Whereas a statement can only be true or false in classical logic, statements in fuzzy logic may be true or false to varying degrees. This enables fuzzy logic to deal with data and rules that are expressed in an imprecise manner using inexact linguistic expressions. Fuzzy inference is the process of formulating the mapping from a given input to an output using fuzzy logic. The mapping then provides a basis from which decisions can be made, or patterns discerned. Fuzzy expert systems are proving to be a powerful tool in business intelligence and decision support. This project concerns the development of a fuzzy expert system for identifying likely student dropouts at Columbus State University (CSU).

According to a report released by the National Center for Public Policy and Higher Education, a low rate of college completion is a key concern in American higher education. According to ACT (the college testing service), the national average freshmen retention rate is 65.7%. From 2005 to 2010, this rate at CSU was 71%. Colleges and universities across the country, including CSU, are investigating this problem of student Retention, Progression and Graduation (RPG) in order to address it more effectively. The main aim of this research project is to build a fuzzy inference based model using a hybrid knowledge extraction process to predict how likely each freshman student will be to drop their program of study at the end of their first semester. Columbus State University database has student Retention, Progression and Graduation (RPG) data dating back to 1998. This historical data is being utilized to develop and evaluate the proposed fuzzy rule-based inferencing system.

Knowledge extraction for the system will be performed using a top down (symbolic) as well as a bottom-up (data-based) approach. In the top-down approach, rules for the fuzzy model will be derived using the traditional knowledge extraction process involving domain expert interviews. Several persons in charge of university departments that have low retention rates will be interviewed to identify parameters that are significant determinants of student success. Fuzzy-rules designed using this knowledge will be weighted appropriately to reflect their level of significance. In the second phase of fuzzy rule derivation, results of data mining performed on student data will be utilized. A feedforward artificial neural network already trained using the student data will be subjected to weight analysis to derive additional rules for the fuzzy rule base, as well as for adjusting the significance of all rules. This hybrid neuro-fuzzy approach is expected to yield better performance than either a conventional fuzzy inferencing system or an artificial neural network.