Detection of Students in Need of Immediate Care in Programming Classes using Machine-Learning Techniques

Abstract

In this study, we propose a novel method for detecting students who need immediate care by the supervisors in a programming class using machine-learning techniques. The system estimates the students’ need for help based on their PC operations including active applications while studying. We implemented a stand-alone application to record the situation. We conducted experiments to investigate the relationship between PC input and the need for help. We found a negative correlation between the number of keystrokes and the need for help, and a positive correlation between the need for help and PDF viewer browsing. The correlation between the supervisor’s decisions and the estimate by DNN was 0.42 (p=0.00), suggesting the possibility of detecting students who need immediate help using a machine-learning approach. Our future work includes to collect meta information and to construct a model to estimate the need for help and evaluate its accuracy in real-time.