Enhancing undergraduate writing skills through blended learning: An artificial neural network model for predictive assessment

The rapid transition of educational institutions toward online and blended modes of instruction has intensified the need for reliable methods to understand and predict students' writing development. This study presents a comprehensive investigation into the enhancement and prediction of writing proficiency among 975 students using Artificial Neural Networks (ANN) and drawing upon recent advances in educational data mining and the proven effectiveness of ANN architectures such as APPNet and SLPNet during the COVID-19 era, a deep learning model that integrates student engagement indicators, demographic variables, and pre-assessment writing scores to forecast final writing performance. The proposed ANN model achieves a prediction accuracy of 92.3%, substantially outperforming traditional machine learning approaches, including Support Vector Machines (84.2%), Logistic Regression (81.7%), Decision Trees (78.5%), and Random Forests (86.9%). Feature analysis reveals that engagement-related factors, particularly revision frequency and peer feedback interaction, exert the most decisive influence on writing outcomes. The findings underscore the pedagogical value of learning analytics in identifying at-risk learners, informing early intervention strategies, and supporting the systematic design of personalized learning pathways to develop students' writing skills in blended learning environments.