An Integration of Unsupervised Approach of Machine Learning in Item Bank Test System

Assessment plays a very important role in schools or universities particularly in evaluating cognitive level of learners. And written assessment such as final examination accounts major component of assessment style in universities. It is a common sight that test or past examination papers along with some pre-defined parameters of test items such as time, question type, knowledge point, difficulty level and others are stored in the repository in the item test bank system. Very often, these exam papers are made accessible to learners to facilitate them for preparing final examination. Over time, there can be a lot of exam papers being accumulated for a particular subject and this may not feasible for learners to go through all of them due to time constraint. A better approach is to form groups or clusters of exam papers according to some attributes such as difficulty level and topics covered so that learners could choose limited number of exam paper to attempt from each group instead of all papers. This could eliminate the need for learners to search and attempt some exam papers that have high similarity and instead exposed to different level of difficulty of exam papers between groups. This requires the use of unsupervised machine learning approach particularly clustering technique to cluster past examination papers automatically according to keywords. These keywords may reflect some pertinent information such as difficulty level, topics covered and so forth. In light of this, this work proposes an enhancement to the existing framework of item test bank system by integrating the facility to cluster past examination papers. In the preliminary testing, the results indicate that clustering can be used to cluster exam papers into groups with a reasonable accuracy.