Abstract
The prediction of students’ performance can be used to discover students who need to be helped in academic, which is an important step towards personalized education. However, majority of studies focus on extracting statistical features manually from the pre-stored data, but the manually features have limited representation capability. And existing studies process missing data by simply removing it, which is not full use of data information. Thus, this paper proposed K-Nearest Neighbor based Time Imputation algorithm for Campus In-and-out Behavior (KNNTI-CIOB) to fill the missing time of in-and-out behavior based on students’ campus behavior regularity. To extract the deeply feature, this paper proposed Hybrid Behavioral Intention Extractor (HBIE) framework. Specifically, the HBIE extracts the long-term and short-term behavior intention by using Transformer Encoder and Gate Recurrent Unit (GRU) to discover the student’s behavior intention. The experiments have been carried out on a real-world dataset. The proposed KNNTI-CIOB algorithm improves the accuracy, recall and precision by 13.1%, 2.82% and 4% respectively in best cases on the same condition by comparing with missing data without processing. The proposed HBIE framework has the best performance in poor academic performance students’ predictions. The accuracy, recall and precision of BIFE are 74.99%, 50.41% and 50.49%, respectively. And the training speed of the classifier is the fastest while the training data is extracted by HBIE.
Funding Information
  • Erasmus
  • National Natural Science Foundation of China