Performance Predictor: A Data Mining and Machine Learning Software for Student Performance Outcomes

Abstract

The intent of this study was to investigate the feasibility of implementing machine learning model for automatically predicting students particularly ICT 1110, who are at risk of failing ICT 1110. Performance predictions as highly important as it is could be very cumbersome as an educator has to analyze large sums of data in order to identify performance especially those at risk of failing so they can administer appropriate correction mechanisms. Over the past 2 years the University of Zambia recorded a high poor performance of students in the ICT 1110 course. There could be many reasons associated with their poor learning outcomes associated with their poor learning outcomes which could be associated to workloads, attendance, lack of resources just name a few. In an attempt to solve this problem, this study presents a performance prediction software that will involve the use of data mining and machine learning in order to train data, associated with student’s academic performance that will be able to predict students who are at risk of failing ICT 1110. The discoveries of the study will benefit the educator as it will predict those students who at risk of failing so that the educators can appropriately execute correction mechanisms to help the students at risk. The research carried out was both quantitative and qualitative. The study targeted a sample size of 60 students (total number of participants) which comprised of the ICT 1110 lecturer, tutor and students. Online interviews and questionnaires were employed to collect data via Google Meet and Google Forms concerning the factors that could possibly be related to student academic performance. The quantitative data was analyzed using Microsoft excel. The results showed a number of factors that could be associated with student performance outcomes which include student interest, mode of teaching, prior knowledge, motivation, support structures, time management, workload, guidance attendance, program minors, participation in course activities and engagement with the course activities. Factors used outside the elicitation process included gender, institutional aid and tuition support. After analyzing the factors to determine the data sources associated with for the factors, it was concluded that student interest, workload, engagement, minors, gender, institutional aid and tuition support would be used as machine learning input features for the model in order for the model to make predictions. In conclusion, the factors analyzed were seen to be effective potential features for the model to identify at risk students accurately in order for educators to render corrective measures to these struggling students.