2021 Fiscal Year Final Research Report
Development of Machine Learning Models for Classifying Physical activity and Sedentary Behavior in Free-Living Toddler
Project/Area Number |
18K13139
|
Research Category |
Grant-in-Aid for Early-Career Scientists
|
Allocation Type | Multi-year Fund |
Review Section |
Basic Section 09030:Childhood and nursery/pre-school education-related
|
Research Institution | Meijo University (2020-2021) Shizuoka Sangyo University (2018-2019) |
Principal Investigator |
|
Project Period (FY) |
2018-04-01 – 2022-03-31
|
Keywords | 幼児 / 身体活動 / 座位行動 / 加速度計 / 行動観察 / 活動タイプ / 活動認識 / 機械学習 |
Outline of Final Research Achievements |
We recorded free-living activities of eleven infants, aged 16-35 months, while wearing accelerometers. Based on these data, we developed and compared models that classify the data into nine activity types using machine learning and deep learning techniques. The results showed that the prediction accuracy was improved by adding gyroscope data, lag window data, and data from hip and wrist worn accelerometer. We also showed the possibility of further improvement in accuracy by using deep learning methods.
|
Free Research Field |
発育発達,健康・スポーツ科学
|
Academic Significance and Societal Importance of the Research Achievements |
本研究では,1~2歳児の身体活動を従来の方法よりも高い精度で判別できるモデルを開発することができた.また,さらにモデルの精度を向上させる可能性のある方法も見えてきた.これらの研究成果は,今後,歩き始めの子どもの身体活動を定量化することを可能にし,この年代の身体活動が,健康や発達に与える影響を明らかにするために役立つ.世界的には歩き始めの子どもの身体活動ガイドラインが策定されている一方で,日本ではそのようなガイドラインはない.1~2歳児の身体活動データが蓄積できれば,日本の身体活動ガイドラインの策定にも貢献することが期待される.
|