Predicting Student Failure using Peer-based Evaluation and Ratings

Detecting student performance in programming is essential for educators to refine their teaching methodologies and for students to gain insight into their strengths and areas needing improvement. This research paper focuses on the accurate identification of students who may struggle or require additional support in understanding the Java programming language. Data was collected through an online questionnaire and an offline exam based on Java, involving 111 Master of Computer Applications (MCA) students. The exam incorporated peer-based evaluation, where students’ answer sheets were assessed by peers from the same class. A total of 29 attributes, including peer evaluation ratings and student behavior survey attributes, were considered for analysis. To identify students requiring additional assistance, five machine learning algorithms, namely Random Forest, Support Vector Machines (SVM), Naive Bayes, K-Nearest Neighbors (KNN), and Decision Tree, were employed. By leveraging various features and patterns within the student data, these algorithms effectively determined which students may need extra support in their Java programming skills. Random Forest yielded the highest overall accuracy of 96.43%. The primary goal of this study is to enable faculty members to concentrate their efforts on supporting and improving the knowledge of students who are at higher risk of failing or require further focus on Java programming. The findings from this research will aid in providing targeted assistance to the identified students, thereby enhancing their understanding and performance in Java programming. The successful implementation of machine learning algorithms for detecting students in need of support contributes to the continuous improvement of student learning outcomes and teaching methodologies.