Research Summary: Revision-Aware Success Prediction from Multi-Attempt Programming Trajectories

This research explores the prediction of programming outcome success using machine learning and deep learning models, including transformer-based pre-trained models. It formulates three distinct prediction tasks based on the current attempt, the next attempt, and success within a three-attempt recovery window, utilizing multi-attempt programming trajectories. The study aims to enhance data-driven programming education through improved learner modeling and adaptive assistance.

Why it matters

Accurate prediction of learner success in complex technical tasks, such as programming, is vital for developing effective educational strategies and support systems. This capability can lead to more efficient resource allocation, personalized learning paths, and ultimately, a higher rate of skill attainment among individuals engaging with technical education.

Key insights

• Programming outcome prediction is critical for data-driven educational approaches.
• Challenges in prediction include diverse error states, short-term revisions, and varying availability of future data.
• Three prediction tasks are defined: current attempt acceptance, next attempt acceptance, and acceptance within a three-attempt recovery window.
• Evaluations cover current-only, pairwise, and multi-step input approaches.
• The study employs Machine Learning (ML), Deep Learning (DL), and transformer-based Pre-trained Models (PTM) for these predictions.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.26169