Neural Network Model for Predicting Students’ Achievement in Blended Courses at the University of Dar es Salaam
Loading...
Date
2017-03
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
International Journal of Artificial Intelligence and Applications (IJAIA)
Abstract
Educator’s knowledge about the likely students’ achievement in blended courses prior to sitting for examinations provides room for early intervention on students’ learning process, especially to those at risk. Unfortunately, Leaning Management Systems (LMSs), Moodle in particular lacks an environment to assist educators access such knowledge from time to time before undertaking their examinations. This raised the need to propose a model, of which from time to time would be providing the likely students’ achievement based on activities in Moodle and previous achievement, taking a case of postgraduate programmes at the University of Dar es Salaam.
This study applied artificial neural networks in building a prediction model. Simulations were conducted in Matrix Laboratory (MATLAB) utilizing seventy eight instances (78) of students’ logs of three blended courses extracted from Moodle for 2013/2014 and 2014/2015 academic years. Mean Square Error (MSE) and Coefficient of Determination (R2) performance metrics were used to find the best prediction model considering ten possible models. The study revealed a model with architecture of 4:10:1 trained with Bayesian Regularization (BR) to be the best model resulting to least MSE of 0.0170 and high R2 of 93 on training. During testing, the model successfully predicted 78% of the students’ achievement with risk and pass status.
Description
Keywords
Artificial Neural Networks, Moodle logs, Blended Learning, Moodle, Learning Management Systems
Citation
DOI: 10.5121/ijaia.2017.8203