Beck, Joseph E.
WORC / Worcester Community Project Center
The problem we are trying to solve is to tweak and improve several neural networks for prediction of student correctness on questions. The approach we adopt to solve the problem is to compare different features (drop rate, number of hidden layers, hidden nodes, with/without autoencoder, optimize function, batch size, fully connected or not) in the neural network so that we can improve AUC (Area under Curve) to get the better prediction results. The results obtained in this research include all the AUC data from all the different neural networks. Our obtained results are great basics of following working on the neural network and succeeding studies in the field of education. The highest accuracy of our single output network reached 87%.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.