AI-based assessment of students’ writing skill progress: Implementing “Artificial neural network modeling” vs. “Time series prediction”

Abstract

This study evaluates two prediction models to assess the students’ progress in second language (L2) writing skills through AI-based applications. The study compares Artificial Neural Networks (ANN) including Multi-Layer Feedforward Networks (MLFN) and Generalized Regression Neural Networks (GRNN), versus time series prediction methods. A total of 34 cases from two groups (an experimental and a control group) participated in a writing pretest, 6 periodic tests and a posttest were analyzed in terms of their writing skill development through different methods. The findings indicated that GRNN without linear predictors for training portion, with mean absolute error (MAE) of 0.807, a root mean squared error (RMSE) of 0.99 and R² = 0.88, outperforming the time series models, provided the most accurate and the closest scores to the posttest indicating that the experimental group improved more than the control group. The GRNN shows better performance with predictor in testing portion of data underscored with R² = 0.94 and RMSE = 0.762. The study emphasizes the importance of applying artificial intelligence in education (AIEd) particularly in English language learning and assessment.