在過去的研究中，發現透過機器學習可以更好的預測學生的學習成效，並可 以透過可解釋人工智慧方式對模型的預測進行解釋。但是不同解釋方式可能對同一筆資料帶來不同的解釋，導致教學團隊無所適從。 本研究根據LBLS-467 資料集，一個記錄大學生在線上學習環境的課程中的 行為日誌與問卷資料，透過機器學習的方法識別高風險與低風險學生。另外，我們進一步使用可解釋人工智慧方法(LIME、SHAP)提取模型預測之解釋，並利用六種評估解釋的評估標準評估解釋的穩定程度(Stability) 與真實程度(Faithfulness)，幫助教師選出更好的解釋，更加瞭解學生目前的學習狀況。 研究結果顯示分類器識別了大多數有風險的學生，也就是說根據LBLS-467 資料集可以找出有風險的學生。透過可解釋人工智慧方法可以得到模型預測的解釋，進一步幫助教師可以了解學生的學習情況。而透過六種評估解釋之評估標準可找出品質較好之解釋，使教學團隊可選出更好的解釋。;In past research, it has been found that machine learning can better predict students′ learning outcomes, and it can provide explanations for the model′s predictions through explainable artificial intelligence methods. However, different explanation methods may yield different interpretations for the same data, leaving teaching teams uncertain. This study is based on the LBLS-467 dataset, which records behavioral logs and questionnaire data of university students in an online learning environment. Using machine learning techniques, high-risk and low-risk students are identified. Additionally, we further utilize explainable artificial intelligence methods (LIME, SHAP) to extract explanations for the model′s predictions. We evaluate the stability and faithfulness of the explanations using six evaluation criteria, helping teachers select better explanations and gain a deeper understanding of students′ current learning status. The research results show that the classifiers identified the majority of at-risk students, meaning that with the LBLS-467 dataset, it is possible to identify students at risk. Through the use of explainable artificial intelligence methods, explanations for the model′s predictions can be obtained, further assisting teachers in understanding students′ learning situations. By utilizing the six evaluation criteria for explaining, high quality explanations can be identified, enabling teaching teams to select better explanations.
