Studies of falls detection algorithm based on support vector machine

Some fall detection systems using inertial-sensor based on threshold algorithm have been proposed so far. But, they all not accurate enough to satisfy patients. In order to improve the performance of falls detection system, a support vector machine (SVM) algorithm was proposed in this paper. Firstly, motion data were collected with a porTable inertial sensing device worn at the patients' waist. Then, five eigenvalues were extracted to get more inherent characteristics. Finally, the SVM classifier was used to mark the suspected falls behaviors, whose parameters were optimized by the particle swarm optimization (PSO) algorithm. The experimental results showed that when distinguishing falls and falls-like activities, the accuracy, false positive rate and false negative rate of the SVM based falls detection algorithm was 97.67%, 4.0% and 0.67% respectively, while it was only 90.33%, 22.67%, 7.33% based on threshold under the same condition. The performance improving of the SVM based falls detection system in this paper is promising in elderly group applications.