Prediction Model Grade Point Average using Backpropagation Neural Network and Multiple Linear Regression

Abstract

Education in the 21st century equips students with knowledge and information and the success of achieving academic achievements during the learning process. Students' academic achievement can be seen from various aspects: the Grade Point Average. So far, efforts to predict GPA have not been made. In fact, if the student's Grade Point Average can be predicted from an early age, the study program can implement a policy to improve graduates' quality and make planning, study escort, and guidance more intensive. Based on this urgency, this study aims to produce a predictive model for the GPA of STMIK Amik Riau students in the odd semester of 2019, using the Backpropagation Neural Network algorithm and Multiple Linear Regression. Backpropagation's architectural model is 8 architectures, and 4-5-1 is the best architectural model with MSE at the time of training = 0.00099965532 and MSE during network validation = 0.0038793 with an epoch of 102 iterations and the resulting accuracy value of 95.24%. Meanwhile, the GPA prediction results, after testing using the Multiple Linear Regression algorithm, obtained an MSE value of 0. 0.27966667%, with a Multiple Correlation coefficient (R) of R = 0.9774925 and a coefficient of determination (R2) = 0.95549159. Thus the prediction of student GPA using MLR is accurate because the value of the coefficient of determination (R2) is close to 1.