Fall Detection System: Signal Analysis in Reducing False Alarms Using Built-in Tri-axial Accelerometer

there are extensive researches conducted to detect falls. However, there are still vulnerable in its accuracy in categorizing and differentiating Activities Daily Living (ADLs) and falls as most of existing system cause false alarm. This research addresses the building of fall detection approach by using built-in tri-axial accelerometer. There are four main phases involved in this research: (1) data acquisition, (2) data processing and filtering, (3) feature extraction and selection and (4) data classification. The raw data of simulated ADLs and falls by participants is collected via built-in-tri-axial accelerometer in smart phone, then automatically send towards the computer via wireless communication. Then, the data is processed and extracted. The proposed algorithms were employed, evaluated and compared in analysis. The findings suggest that ANN method with 35 hidden neurons is the most accurate model for fall detection system in this research as it achieved 99.24% overall accuracy while producing 0.18% FPR. This approach has the potential to be implemented and deploy in real-time mobile application in future.