Optimal clustering algorithm for knowledge discovery in University of Ibadan post unified tertiary matriculation examination

Abstract

This study evaluates the performance of five clustering algorithms on the University of Ibadan Post Unified Tertiary Matriculation Examination (PUTME) data. This exercise generates much data which when explored for unknown patterns could lead to knowledge discovery that can be used for planning and programme-execution. The goal is to explore and analyze this data for unknown pattern. However, choosing the most appropriate clustering algorithm can be difficult. Most of the algorithms generally assume some implicit structure in the dataset. Discovering the optimal model inherent in the algorithm would be informative to stakeholders; therefore, the study aims at developing an optimal model for knowledge discovery that would enhance decision making. The evaluation was done by preprocessing, normalizing and feeding the dataset into the various algorithms to create models which were validated, evaluated and stored. The performance evaluation results were analysed, compared and plotted for inference. This varies with the internal parameters inherent in the algorithm, the tuple size and the performance evaluation metric used, but in the average, MeanShift algorithm performed best. MeanShift algorithm is therefore recommended. It was also observed that the age group (16-17) dominated the cluster which represents the successful candidates in Unified Tertiary Matriculation Examination (UTME) and PUTME: the best time for a student to write these examinations. No age 14 was found in the cluster, which implies that no candidate of this age sailed through the examinations. Female candidates from the age of 40 do not most often pursue their education through examinations; some who were determined had a very high percentage success as compared to their male counterpart.