A Empirical Study of Programming Behaviours on Large Scale Online Learning

Traditional teaching methods for programming mainly pay attention to the final artifacts but ignore the process such as the history of test failures and code change. In recent years, modern online programming learning platforms can record the student's behavioral data, which are drawing more and more attention by higher education institutions. Analyzing behavioral data in the modern learning systems can help users understand the students' behavioral patterns in order to design good teaching programs. In this paper, we first focus on the relationship between the time of completing a course and the average test time of fixing test failures. Using the K-means algorithm, we then cluster the error types into four categories based on three factors. The error types can help students master their own error patterns and provide guidance for teachers to grasp the knowledge blindness of the students.