A RISK CLASSIFICATION SYSTEM FOR ELDERLY FALLS USING SUPPORT VECTOR MACHINE

Abstract

Falls are a multi-factor problem that poses a serious risk to the elderly. Approximately, 60% of falls are caused by a number of known factors, including the environment, which accounts for approximately 25–45% of falling risk. Most of the remainder results from a lack of personal balance control. Falling can cause long-term disabilities in the elderly, sometimes resulting in lower quality of life, and is also associated with increased medical expenses and personal care costs. In this study, we developed a falling assessment system to evaluate and classify individuals into four graded falling risk groups. During the test, all subjects were required to wear a self-developed dynamic measurement system and to perform two balance tests: a “Timed Up and Go Test” and a “30-Second Chair Stand Test.” We obtained 29 characteristic parameters from the data recorded during these tests. Next, we performed group classification. Eigenvalues were normalized, and a principal component analysis (PCA) was performed. After identifying informative characteristic parameters, support vector machine (SVM) was used to classify individuals as members of one of the four falling risk groups. These included low-, moderate-, high-, and extreme-risk groups. Using unreduced data of the 29 characteristic parameters extracted from the two balance tests, the accuracy of the SVM classification in allocating individuals to the correct group was 97.5%. After PCA, the 29 characteristic parameters were reduced to eight principal components, and the SVM classification method using these eight principal components was 93.25%.