Applying Clustering Algorithm to Find Important 3MTUG Features of Rehabilitation Patients

• Main Authors: Claudia Putri, 陳美香
• Other Authors: Tien-Lung Sun
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/5g36x6

碩士 === 元智大學 === 工業工程與管理學系 === 107 === In recent years, stroke has moved from second to the fourth leading cause of death, it attacking elderly and young people. In additions most of the stroke people have a great risk in the fall, therefore conducted various studies to prevent the risk of falling over on the stroke people such as rehabilitation to cure their stroke. It collaborated with hospitals to took an assessment by measuring a stroke in-patient balance as timed up and go, Short Form Berg Balance Scale (SFBBS), the physical performance test, etc. Segmentation the data: Sit to stand, stand to walk, turn, walk back and walk to sit is the part of important to understand the data and to do process analysis using unsupervised algorithm (t-SNE). T-SNE helping reducing the multi-dimensional into two or three dimensional with using parameters: perplexity, projections and iterations. The result of t-SNE is there is relationship between TUG and SFBBS with have cut off is greater than equal 20 for stoke people since in scatter plot have a little outlier, there is blue points in red cluster, it could happen maybe they afraid of falling so they TUG result in red cluster and red points in blue cluster it could happen since they using auxiliary to prevent of falling. By looking the first t-test result, the t-SNE result and the last t-test result that the important segment and direction is ML-axis of Whole segment and ML-axis of Sit to Stand segment which means this features is important to analyze the balance of stroke in-patient. From the SFBBS observation is task #7 is the hardest part to do to stroke patient. T-SNE is very useful to clustering analysis that can help to identify outlier and define a cluster.