Mining Acceleration Data for Smartphone-based Fall Detection

Abstract

Falls have become a public health problem that directly affects the quality of life in the elderly people. For many years, a variety of methods has been studied to develop a system that detects falls at the earliest, aiming to avoid their consequences. However, existing systems developed for monitoring a person of interest are expensive and impractical to use. Recently, smartphone-based fall detection systems that make use of a built-in accelerometer sensor have been proposed to address the above limitation. However, high false alarm rate signir1cantly limits the effectiveness of smartphone-based fall monitoring. In this work, we propose a novel data mining algorithm for fall monitoring. It discovers sequence patterns from data of the accelerometer and then utilizes the extracted patterns for building a reliable fall detector system on mobile platform. We conduct experiments on Mobifall and real datasets for evaluating the proposed method. The experimental results conr1rm that our method achieves a high detection rate with acceptable false alarm ratio compared with state-of-the-art smartphone-based fall detection algorithms. a variety of methods has been studied to develop a system that detects falls at the earliest, aiming to avoid their consequences. However, existing systems developed for monitoring a person of interest are expensive and impractical to use. Recently, smartphone-based fall detection systems that make use of a builtin accelerometer sensor have been proposed to address the above limitation. However, high false alarm rate significantly limits the effectiveness of smartphone-based fall monitoring. In this work, we propose a novel data mining algorithm for fall monitoring. It discovers sequence patterns from data of the accelerometer and then utilizes the extracted patterns for building a reliable fall detector system on mobile platform. We conduct experiments on Mobifall and real datasets for evaluating the proposed method. The experimental results confirm that our method achieves a high detection rate with acceptable false alarm ratio compared with state-of-the-art smartphone-based fall detection algorithms.