Abstract
In this study, we analyzed the features for judging whether a player understands the program he/she is typing based on keystroke information such as typed keys and their times in a program typing game. The results showed that (1) the average typing speed during a typing game can be used to separate the understanding from the non-understanding with about 70% accuracy, (2) a machine learning model using the relative interval time normalized from 0 to 1 for each individual can be used to classify with about 80% accuracy, and (3) the features that machine learning discovered were that people who understand the program type line breaks relatively slower than other keys. We further analyzed whether these characteristics appeared as understanding of the program increased, or whether they were inherent characteristics of the user, and found that they appeared from the beginning of the program class. We can infer that some kind of programming background influences the keystrokes.
https://ror.org/03nskz175 
