Personalized prediction of student progress in moral education using multi modal learning

In modern education systems, accurate and individualized prediction of the development of students in ethics education is the central problem. Unbiased insight into the moral development path of each individual, and the provision of appropriate and timely interventions, are essential for the promotion of social well-being as well as the education of civic citizens. The present study aims to propose a hybrid method for individualized prediction of the development of students in ethics education. The proposed method consists of four key steps: (1) Preprocessing raw data to prepare data for the subsequent steps; (2) Modeling data modalities and encoding them into Auto-Encoder neural networks, where it is possible to extract meaningful representations of various aspects of data; (3) Fusion and aggregation of created modalities through a self-attention mechanism so that the model can actively choose the importance of every modality to forecast the target variable; and (4) Target variable forecasting based on a random forest by utilizing the set of fused features from previous steps. This new approach, founded on innovative machine learning techniques and data fusion, attempts to provide more accurate and reliable predictions about the path of students' moral growth. These results prove the effectiveness of the proposed approach and show improvements in terms of accuracy by 7.1% and F-measure by 7% compared to competing methods.