Implementasi Algoritma Sbert dan XGBRegressor untuk Penilaian Otomatis dan Pendeteksi Plagiarisme pada Ujian Esai

Abstract

The development of technology has brought significant changes to the field of education, particularly with the emergence of online exam applications and websites. Online exams consisting of multiple-choice and essay types have their advantages and disadvantages. Essay exams can measure students' analytical and expressive skills in their own words and reduce the likelihood of guessing answers, but they have disadvantages in terms of assessment reliability, long assessment time, and potential cheating among students. From these problems, a system is needed that can help check the exam from the assessment and plagiarism detection automatically and accurately. This research utilizes SBERT and XGBRegressor algorithms, which are well-suited for handling data in sentence form. The data used is a collection of essay exam questions and answers. This dataset was collected from junior and senior high schools in Letjen S. Parman, Medan, with a total of 550 answers for 35 types of questions. The dataset is then divided into 80% for training data and 20% for testing data. Then, the training data is used to train the XGBRegressor model to assess students' answers to the teacher's answer key and detect plagiarism among students. Then the model will be tested with test data to measure the accuracy of the model's prediction results. The accuracy obtained from testing the assessment model is MAE (Mean Absolute Error) of 1.5500, RMSE (Root Mean Squared Error) of 2.9267, R2 Score of 0.9747, and Pearson Correlation of 0.9886. The accuracy obtained from testing the plagiarism detection model is MAE (Mean Absolute Error) of 1.1298, RMSE (Root Mean Squared Error) of 2.1786, R2 Score of 0.9921, and Pearson Correlation of 0.9961. The test results show the reliability of the model in accurately representing the original value.