Chi-Square Oversampling to Improve Dropout Prediction Performance in Massive Open Online Courses

Massive Open Online Courses (MOOCs) are important to achieve educational quality in Indonesia. However, low retention rates are global problems that must be addressed by building a prediction model to prevent dropout. The prediction model faces a challenge due to the disproportionate comparison between major and minor data. In this study, the 141 datasets collected from the questionnaire consisted of 95 participant data who completed the course and 5 dropout data. This necessitated oversampling to balance the data using Synthetic Minority Over-sampling Technique for Nominal (SMOTE-N) and SMOTE for Encoded Nominal and Continuous (SMOTE-ENC) chi-square methods. The dataset formed was processed using Support Vector Machine (SVM) machine learning method. In the testing process, the performance of the prediction model with SMOTE-N and SMOTE-ENC chi-square oversampling data was compared with the prediction model with regular oversampling data. The results showed a significant increase in accuracy from each oversampling method with weighting. SMOTE-N weighting modification using chi-square value had the best value, with F1-measure reaching 95.33, and a decrease in error in the prediction of dropout data was observed. This result showed that the model formed with the SMOTE-N chi-square method has good predictive ability.