Student placement predictor for programming class using classes attitude, psychological scale, and code metrics

It is often necessary to divide a class according to students' skill level and motivation to learn. This process is burdensome for teachers because they must prepare, implement, and evaluation a placement examination. This paper tries to predict the placement results via machine learning from some materials without such an examination. The explanatory variables are 1. Psychological Scale, 2. Programming Task, and 3. Student-answered Questionnaire. The participants are university students enrolled in a Java programming class. The target variable is the placement result based on an examination by a teacher of the class. Our classification model with Decision Tree has an F-measure of 0.937. We found that the set of the following explanatory variables can yield the best F-measure (0.937): (1) Class Fan Out Complexity, (2) Practical utility value, (3) Difficulty Level 4 (AOJ), (4) Difficulty Level 3 (AOJ), (5) Interest value, and (6) Never-Give-Up Attitude.