Abstract:

Currently, the number of Japanese higher-education organizations, including universities and junior colleges, is over 1,000. The total number of private universities is about four times the number of national and public universities. The number of higher-education universities is currently too high in Japan. Due to declining birth rates in Japan, 40% of private universities have not been filled up to their capacity. Students of these private universities are often not attending their first-choice schools, and therefore, their motivation to study is generally low. Because of this, these private university students often fail to graduate, resulting in high dropout rates.

In order to address this problem, most Japanese universities are offering a tutoring program for these students. The tutors are generally chosen from among the high-achieving senior and graduate students, but it is difficult to secure a sufficient number of tutors at these universities. At some universities, teachers may also serve as tutors and mentors for the students. However, it is difficult for a teacher, who already has many responsibilities to fulfill in their primary role, to also fulfill the role of a tutor. One method to alleviate the strain on the universities’ tutoring programs, would be to detect abnormal behavior of students based on attendance status and their own learning history, and provide appropriate instruction to the students via their tutors, thus making tutoring more effective, and to prevent student dropouts in advance.

In this research, we propose a tutoring assistance framework for teachers. Machine learning is applicated to determine the student's learning history, their class attendance status, and the student's accumulated academic credit acquisition status and dropout risk. If it is determined that a student's possibility of dropout is increasing, the teacher receives a notification via the framework. Once the teacher notices the existence of students who have any problems, the teacher can provide remedial instruction directly to help the declining student . As a result, students' problems can be solved at an early stage, and the possibility of dropout is reduced. The proposed framework does not support detected students directly, rather, it provides awareness of class management for teachers who are also tutors.

This paper first describes the current situation of Japanese higher educational environment, especially at private universities, and the problem of the difficulty of keeping a student's motivation high. Second, we describe the related studies addressing the use of the student assistance framework for maintaining their motivation to study, and method for supporting tutors. Third, we present the design of the proposed assistance framework for tutors using data from student learning history and machine learning analytics. Finally, we describe the results of the experimental use of the prototype of the proposed assistance framework, and we present the effectiveness of tutoring based on the use of the assistance framework.

Keywords:

Tutoring assistance, learning analysis, abnormal behavior detection, machine learning.