With the ageing of the global population, fall detection of elderly become a prominent public health problem. Health service, academia and industry are desirous to develop a robust system for automatic falls detection in elderly's daily life, especially for the elderly who living alone. In this paper, we develop an optimize falls detection algorithm that initially investigates the problems of existing systems, and then collect numerous activities of daily living (ADLs) and fall events dataset by a self-defined method. The algorithm is deployed within a sensor system that uses a triaxial accelerometer and a triaxial gyroscope sensors to detect accidental falls. The aim of this paper is to explore the favorable body location for single-sensor wearable falls detection device to detect the falls for elderly. The developed sensor system is used to experiment on seven different locations of human body. The result is detailed evaluated by measuring sensitivity and specificity of algorithm applied in each experimented sensor locations.