Adaptive time method for fall detection in elderly

Falls are the largest risk to the health of elderly people. Different researchers have studied threshold algorithms and machine learning for fall detection; however, these methods do not efficiently detect the possibility of falls in the elderly. Recently, sensors have been developed that can observe human joints to determine how prone a person is to fall. However, the use of a single sensor exhibits several drawbacks such as low accuracy, limited information, and a high false alarm rate. Therefore, multiple sensors at the waist, thigh, and ankle were the most comfortable positions for elderly people wearing sensors. In this paper, a hybrid approach to dimension reduction and discrete wavelet transform (DWT) is proposed to extract features from the dataset. A unique approach employing k Nearest Neighbor (KNN) and Support Vector Machine (SVM) was investigated to accurately detect if a person is prone to fall or not. The results after comparative analysis with current methods show a significant increase in accuracy of 94%.