Student placement and skill ranking predictors for programming classes using class attitude, psychological scales,...

Area of Science:

• Computer Science
• Educational Technology
• Machine Learning

Background:

• Assessing programming skills is crucial for student placement and employee recruitment.
• Current methods involving programming tests and contests are labor-intensive for educators and recruiters.

Purpose of the Study:

• To predict programming contest placement and ranking outcomes using machine learning models.
• To reduce the burden of manual evaluation in programming skill assessment.

Main Methods:

• Machine learning models, including decision trees and SVM-rank, were employed.
• Explanatory variables included Psychological Scales, Programming Tasks (e.g., source code complexity), and Student-answered Questionnaires.
• Participants were university students in a Java programming class.

Main Results:

• The best classification model (decision tree) achieved an F-measure of 0.912.
• The best ranking model (SVM-rank) achieved an nDCG of 0.962.
• Programming Tasks were the most significant explanatory variables, followed by Psychological Scales and Questionnaires.

Conclusions:

• Machine learning can accurately predict programming contest results, automating evaluations and potentially enhancing educational quality.
• Difficult-to-quantify data, such as psychological scales, can be effectively utilized in skill assessment.