An Exact and Inexact Data Matching Approach for Saving Time and Preventing Errors in Processing of Student Exam Results at the University of Botswana

Abstract

The University of Botswana (UB) offers a comprehensive range of academic programmes, leading to a variety of undergraduate and postgraduate qualifications offered across its seven faculties. Like other universities, UB faces a problem of large enrolment creating a large numbers of students in some of the courses offered. Often this problem is seen in the university-wide General Education Courses which UB created to provide basic interdisciplinary skills for students. We study an examination problem using one such course, Computing and Information Skills Fundamentals I, with a registration of more than 4000 students. The exam questions are designed with multiple choice questions and the exam answer sheets are marked using an Optical Mark Recognition scanner. A recurring problem is observed where students do not shade their details perfectly, leading to a cumbersome problem when matching exam records with class list records. We implement a data matching solution and test it, yielding excellent results. We make use of an exact matching processing step, before use of standard data matching methods for indexing, comparison, and classification. We use q-gram indexing, edit-distance comparison, and Support Vector Machine classification methods. The results show that the impact is very high. First the processing time is reduced from a cumbersome process of more than 13 hours to less than 18 minutes, with no mistakes made in the automated processing phase. Secondly, record pairs which had too little information to match automatically and have to be matched manually are few hence manageable