Empirical Study Of Elderly Plantar Pressure Feature Selection And Predictors Of Falls Using KNN-based Algorithms

Fall is a common and serious problem for the elderly people in daily life. The occurrence of fall isn't taking place by chance. There is a potential risk factor that can be prevented and controlled. The early fall risk assessment method is main strategy in evaluating fall risk of highrisk patients, so as to be targeted interventions for this population. This paper takes elderly people as experimental subjects, and obtains their plantar pressure data while taking nomal walking and sit-to-stand. We aimed to determine whether objective measures of physical function could predict subsequent falls, and designed a fall risk assessment system with plantar pressure sensors using KNN-based algorithms to predict fall.MethodsThis paper took the community elderly whom belong to Beijing City Malianwa Street as the experimental objects. A total of 38 participants were included and aged above 65 years old. All participants were categorized as fallers or non-fallers, according to the falling experience and the scores obtained at the balance ability test. This essay applied plantar pressure distribution system to obtain thenar dynamic parameter of each left and right foot during walking or sit-to-stand. Sample entropy was used to quantify variables of plantar pressure. A feature selection algorithm was used to select relevant features to classify the elderly into two groups: at risk and not risk for falling down, for three KNN-based classifier: local mean-based k-nearest neighbor(LMKNN), pseudo nearest neighbor(PNN), and local mean pseudo nearest neighbor(LMPNN) classification. We compared classification performances, and selected optimal feature sets with best results.Results(1) Sample entropy of plantar pressure of two groups was analyzed. The statistical characteristics of the corresponding variables differ between left and right foot. Sample entropies of plantar pressure for right foot were more significantly different among two groups.(2) We compared and analyzed classification performances, and achieved the best results with LMPNN, with sensitivity, specificity and accuracy all 100%. The classification results further indicated that L_SI_F; R_ML_F; R_AP_F; L_V_F could differentiate persons who had previous fall event.Conclusions(1) For the first time we applied sample entropy into analyzing time series signals of plantar pressure. The statistical characteristics of the left and right foot were significantly different. Therefore the data sets of plantar pressure should be collected and analyzed dependently between the left foot and right foot.(2) A automatic classification system based plantar pressure was implemented. Optimal feature set and classification parameter are important for the classification performance. The experimental results include classification result and selected feature set. This is the first step in the design of objective fall risk assessment system that could be useful in evaluating balance and the risk of falling down.