Evaluation of classification algorithms for predicting students’ learning performance based on bookroll reading logs

To help students’ learning, finding at-risk students is an important issue in education. This study has collected students’ learning logs in ebook reading environment for 17 blended learning courses in Taiwan. Then, we have construct students’ learning performance prediction model based on learning logs by using eight classification methods. To improve prediction performance, we have also propose grading policy rules which include of grading on stringency rule, grading on moderate rule, and grading on leniency rule to define the labels of students’ learning performance. According to the prediction performance results, the LR is the suitable classification method for predicting students’ learning performance. Besides, the grading on leniency rule has obtained the higher prediction performance. For exploring the influence factors on classification performance, we investigate the relationship between students’ online learning actions and students’ academic performance by Spearman coefficient analysis the open file (f1), the number of delete marker (f6), the number of next pages (f12), and the number of previous pages (f13) features are the critical factors for affecting the prediction performance of learning logs.