A Machine Learning Based Auto-Grading Model for Question-Based Algebra Problems using RoBERTa

The use of Machine learning models into educational assessment has revolutionized the grading process for problemsolving domains.This research introduces a machine learning auto-grading model tailored for question-based algebra problems, utilizing RoBERTa, a natural language processing model.The primary objective is to investigate the adaptability of RoBERTa, originally designed for natural language understanding, for representing non-natural language mathematical expressions. The study also explores the impact of different embedding strategies on the performance of autograding systems for student responses to equation problems.The mathematical and textual code inputs were transformed into vectors, and the vectors were then fed into several models. Evaluation metrics such as RMSE, MSE, R2, and MAE are utilized to assess the effectiveness of each embedding approach in grading student responses accurately. The use of k-fold cross-validation will improve the dependability of the assessment measures, offering insights into the model’s consistency across different subsets of the data. The comparative analysis of different embedding strategies offers guidance on selecting the most suitable approach for enhancing the autograding process.