ANALYSIS FOR CHARACTERISTICS OF LOW-ACHIEVERS AND CONSTRUCTION OF A CLASSIFICATION MODEL USING PROGRAMMING DATA

Since 2018, programming has become a compulsory component of information education in Japanese senior high schools.Programming proficiency should be increased by improving instruction for difficult to understand content.This study aims to construct a model to classify students into subgroups at risk of low proficiency using machine learning, focusing on the types and number of token occurrences in student-created programs.Students created programs using a newly developed web-based coding system.When an operational event such as character input or execution occurs, the type of event, time, and current program are sequentially recorded as programming data.The student-created programs were divided into tokens and the occurrences of these tokens was analyzed.A classification model was constructed using machine learning to detect subgroups based on the number of occurrences of each token.It was found that low achievers could be classified with an accuracy of approximately 70% using this model.This has important implications for providing focused instruction for weaker students.