Authors : N. Sharada, M. Shashi and Xiaolu Xiong
Abstract: With accurate and dependable student knowledge retention models both over practice and under practice can be avoided in adaptive educational systems. The knowledge retention being a hidden parameter and hence can only be estimated from student responses to the retention tests conducted over a period of time. While the relationships among the independent variables describing the student experiences is nonlinear and complex, the existing personalized scheduling systems attempted to model them using traditional methods such as linear regression and basic statistical techniques. In this research, the application of the most advanced computational techniques such as deep learning and random forest are investigated on the dataset from Personalized Adaptive Scheduling System (PASS) module of ASSISTment, web based mathematics tutor. Experiments demonstrate that deep learning technique for student knowledge retention modeling significantly outperformed the baseline GLM Model with an R2 value of 0.542. In addition, these techniques are further explored to schedule personalized retention tests after the students initial skill mastery. For this regression problem, the random forest regression technique, indicated a prediction improvement with an R2 of 0.912 than a baseline linear regression model with an R2 of 0.417.
N. Sharada, M. Shashi and Xiaolu Xiong, 2018. Modeling Student Knowledge Retention Using Deep Learning and Random Forests. Journal of Engineering and Applied Sciences, 13: 1347-1353.