Fumiko Harada

Work place: Research organization of Science and Technology, Ritsumeikan University, Japan

E-mail: harada@fc.ritsumei.ac.jp

Website:

Research Interests: Computational Science and Engineering, Computational Engineering, Data Structures and Algorithms

Biography

Fumiko Harada received the PhD at Graduate school of Engineering science, Osaka University. She was working as the assistant professor and lecturer at College of Information Science and Engineering, Ritsumeikan University during 2007-2009 and 2009-2012, respectively. She is currently a visiting professor at Research Organization of Science and Technology, Ritsumeikan University. Her research interest includes data engineering, IoH applications and recommendation systems.

Author Articles
Predicting Online Student Effort with Accelerometer, Heart Rate Sensors, and Camera Using Random Forest Regression Model

By Fumiko Harada Rin Nagai Hiromitsu Shimakawa

DOI: https://doi.org/10.5815/ijmecs.2022.05.02, Pub. Date: 8 Oct. 2022

In online education through web conference tools, teachers cannot grasp students' states by watching their behaviors like in an offline classroom. Each student also cannot be affected by others' good behavior. This paper proposes a prediction method of the student effort through acceleration sensors and a heart rate sensor worn on a student's body, and a local camera. The effort is expressed by the levels of concentration, excitation, and bodily action. A Random Forest regression model is used to predict each level from the sensor and camera data. Exhibiting the prediction result brings visibility of student states like offline. We verified the effectiveness of the prediction model through an experiment. We built the Random Forest regression prediction models from the sensors, camera, and student effort data obtained by actual lectures. In the case of building one prediction model for one lecture/one subject, the average R2 values were 0.953, 0.925, and 0.930 in the concentration, excitation, and bodily action, respectively. The R2 was -0.835 when one prediction model trained by one lecture's data is applied for another lecture's prediction. That was 0.285 when one model by 4 subjects' data is applied for prediction for the rest 1 subject. It means that the prediction model has high accuracy but is dependent on individual persons and lectures, which forces a burden to individual student to collect initial training data for individual lecture to build a prediction model. We also found that the acceleration data are the most important features. It implies the effectiveness of using acceleration sensors to predict student effort.

[...] Read more.
Other Articles