Predicting the Students Performance using Regularization-based Linear Regression

The corona pandemic changes the education paradigm from offline- to online learning. This situation causes a crucial problem in the evaluation process of student outcome learning. A valid enough evaluation is difficult to achieve because of the lack of face-to-face between teachers and students. However, this stage is critical in the education and teaching process. Therefore, our research focuses on building a prediction model of students' performance for supporting the teacher in the evaluation process. The prediction model is created by linear regression based on Regularization. Further, we explore three regularizations, namely: ridge, lasso, and elastic net, to find the best performance of the model. The evaluation technique used is random sampling with various training set sizes. In addition, we determine the different values of alpha to discover the best model. The experimental results show that the ridge regularization model's prediction error rate is lower than the lasso and elastic net Regularization. The results of measuring values using MSE, RMSE, MAE, and R2 show that the prediction model built using ridge regularization is superior to the others.