Assessing The Importance of Dynamic Features of Academic Performance Using Machine Learning Models During Distance Education Caused By COVID-19

In this work, we perform an analysis of the impact of transition to the distance learning format caused due to the COVID-19. During the critical period of the pandemic, profound transformations have transpired in the sphere of education, necessitating a reconfiguration of the traditional teaching and learning methods. Due to the imperative of ensuring safety and minimizing the spread of the virus, students have been compelled to refrain from physical attendance at educational institutions, leading to their separation from teachers and peers. Current study is based on an extensive dataset, containing the academic grades of school students during 2016-2020 academic years, extracted from the database "Electronic Education in the Republic of Tatarstan". Explainable artificial intelligence (XAI) method is commonly applied for estimation feature importance, allowing the evaluation of each analyzed feature impact on the output of machine and deep learning models. The data analysis and interpretations were performed by using machine learning and deep learning algorithms, and by using modern XAI framework SHAP, on the aggerated datasets containing quarter’s and month’s mean grades for 3 consecutive years for 4 school subjects. As a result, we measured subjects’ importance for quarters and months scale that make the greatest contribution to the forecast of academic performance after distance learning beginning.