Classification of Falling Risk using Support Vector Machine based on Quantitative Falling Assessment System
Falling is a multi-factor problem that easily occurs in the elderly, and about 60% of which is caused by a variety of reasons. The environmental factors account for about 25- 45% of the causes of falling. Most of the rest is due to lack of personal balance control. Falling can cause long-term disabilities in the elderly and lower the quality of life. It will also pay a large amount of medical expenses and human resources. In this study, the self-developed falling assessment system was applied to evaluate and classify four types of falling risk groups. During the period of the case record, all subjects were required to wear the self-developed quantitative falling assessment system to do two kinds of evaluation balance ability test which are "Timed Up and Go Test" and " The 30-Second Chair Stand Test". After analysis, the characteristic parameters were obtained. In the group classification task, the eigenvalues were normalized and the principal component analysis was applied. The important characteristic parameters were therefore observed, and the support vector machine (SVM) were used to classify four types of falling risk groups, i.e., the low-risk group which has never fallen and the function of physical activity is normal, the moderate risk group which has not fallen but the function of physical activity has begun to degenerate, the highly dangerous ethnic group which once had a falling and did not have a good balance when standing or walking, and the extremely dangerous group which had fallen many times. At current stages, four types of falling data were used to test the applicability of the classification system. 29 characteristic parameters extracted from "Timed Up and Go Test" and "The 30-Second Chair Stand Test" were used for SVM classification. The classification accuracy reached 99.15%. After PCA analysis, the 29 characteristic parameters of "The 30-Second Chair Stand Test" and "Timed Up and Go Test" were reduced to only 9 principal components. After SVM classification, the accuracy achieved 96.15%.
Keywords - Falling, Balance Control, Support Vector Machine (SVM).