설명 가능한 인공지능을 활용한 청소년의 학업열의 관련 주요 변수 탐색

Abstract

Recently, the importance of academic engagement, an important key to overcoming under-achievement and academic-related stress, has been highlighted. The purpose of this study was to explore key variables related to adolescents’ academic engagement. For this purpose, data from the 1st year (7th graders) and 4th year (10th graders) of the Korean Children and Youth Panel Survey 2018 were used. A comparison was made of the predictive performance of random forest, gradient boosting, and XGBoost to select the best machine learning technique for predicting academic engagement. Additionally, the SHAP index, which is an explainable artificial intelligence technique, was used to identify the relationship between key variables and academic engagement. The main research results are as follows. It was found that gradient boosting had the best prediction performance among the three machine learning techniques compared. The key variables commonly derived from middle and high schools are self-study time (weekends, weekdays), grade level for all subjects the previous semester, life satisfaction, grit, creative personality (cleverness), and teacher-student relationship, which had a positive relationship with academic engagement. However, academic helplessness (lack of learning motivation, lack of positive emotions) was found to have a negative relationship with academic engagement. The commonly derived key variables were in the areas of daily living time, intellectual development, social/emotional, and school. In addition, major variables were derived in the activity area for middle school and in physical development, media, career, and family areas for high school. Based on these results, educational implications regarding adolescents’ academic engagement are discussed.