Classification of student performance based on first half-semester of online learning using fuzzy K-nearest neighbor

During the Covid-19 pandemic, many face-to-face activities must be carried out online, including the teaching-learning process in higher education. Several problems occur in learning process, especially for students and lecturers who were previously unfamiliar with online learning. This leads to difficulties for students in achieving the learning outcomes. One strategy to detect student success in a lecture is to monitor student achievement from the beginning of the lecture to midterm. In this research, a machine learning method, the Fuzzy K-Nearest Neighbor (Fuzzy KNN) method, is used to classify student performance at the end of the lecture. The student performance is classified into good and poor performance based on student activity data recorded in the Learning Management Systems (LMS). The model developed uses data from the first half of the semester from course X at the Universitas Indonesia which was held during the 2020/2021 odd semester. In order to overcome the imbalance of data in the training data, the SMOTE (Synthetic Minority Oversampling Technique) process is applied. To measure the performance of the Fuzzy-KNN method in the model built, a measurement of accuracy and recall value will be used. In addition, performance of the Fuzzy-KNN method in the model built is also compared with the KNN method. The result shows that the Fuzzy-KNN method yields better results compared to KNN method. The best performance of the Fuzzy-KNN method can be seen from the recall value, where the best recall value can reach 53.65% with an accuracy of 76.64% if using the k-fold cross-validation evaluation method. If data is separated into 80% training data and 20% testing data up to ten trials, the best recall value of Fuzzy-KNN is 51.07% with an accuracy of 77.02%.