Implementation of CRISP-DM to Predict Student Graduation on Time Using Naïve Bayes Algorithm

The problem of on-time graduation of students at Telkom University, majoring in Information Systems, shows a decrease in the percentage of graduation until 2018. The inability of students to complete their studies within the specified time can have an impact on the reputation of the institution, increase education costs and living costs, delay professional careers, and overcapacity of campus facilities. To overcome this problem, a graduation prediction system that is fast, accurate, and effective is needed. This research applies the CRISP-DM methodology combined with data mining techniques using the Naïve Bayes classification algorithm to predict timely graduation. The CRISP-DM approach guides each step of the process, from understanding the problem and preparing the data to model evaluation, ensuring a structured and systematic workflow. The model without SMOTE shows good performance with an accuracy of 86.38%, precision of 95.76%, and recall of 87.78%, and the accuracy varies from 76.1% to 91.5% using K-fold CV with 5 fold experiments. Meanwhile, the model with SMOTE applied resulted in 83.93% accuracy, 94.48% precision, and 78.08% recall, with accuracy varying from 65.9% to 87.8%. Although SMOTE helps in some folds, the model without SMOTE shows a more stable and better performance in predicting timely graduation. In conclusion, the Naïve Bayes model without SMOTE is more recommended to be used in predicting students' on-time graduation.