Using Machine Learning Method to Build a Detecting Physical Conditions System with Postural Balance

• Main Authors: Yu-Che Chen, 陳宇徹
• Other Authors: Chin-Mei Chou
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/tu6c44

碩士 === 元智大學 === 工業工程與管理學系 === 107 === The aim of this study was to build a detecting physical conditions system with postural balance by using machine learning method. The previous research showed the postural balance be related with physical conditions. And some research even used machine learning method to build a predict model for detecting the disease or falling risk. Above of results were well but there were few researches bulding a detecting physical conditions system through the various physical value and postural balance. Therefore, if we could build this system sucessfully, it might be a tool for people to detect physical conditions in the future. The whole research had three part. First, we made a evluation between the force plat and the wii balance board that they had the same center of pressure (COP) value in consistency and reliability or not. Second, we used machine learninig method to build a predict model with postural balance. The process was discussing the Synthetic Minority Over-sampling Technique (SMOTE), the two standing time (30s and 60s) and the different machine learning method (logistic regression, support vector machine, ramdom forest and one-class support vector machine). And the last, we created the system to connect hardware and software so users could operate it to make an examination. The evluation results got the substantial level above and no special image in Bland-Altman difference plot so we could say that they were consistence and reliable. And the model result showed the ramdom forest mothed were the best classifer. Howereve, it took too much time so we choose logistic regression method and the 30s standing time as our predict model. In conclusion, we had a good outcome and suceed to create the system in our research.