Comparison of Machine Learning Algorithms for Position-Oriented Human Fall Detection

Smart sensor systems are increasingly pervading all kind of application fields such as in industry, ambient assisted living, or lifestyle accessories. In this work, a smart system for position-oriented human fall detection is investigated using various machine-learning algorithms for data processing and evaluation. Data from an inertial measurement unit is combined with data from visible light positioning methods to achieve position-based fall detection. Furthermore, an experimental setup and test methods were created to generate appropriate datasets for this analysis. The classification accuracy is compared with three machine-learning algorithms commonly used for such tasks, which are Decision Tree, Naïve Bayes and Support Vector Machine. It is demonstrated that the combination of data from the two sensor systems can improve the recognition accuracy beyond 99% in the best case.