A Machine Learning Multi-Class Approach for Fall Detection Systems Based on Wearable Sensors with a Study on Sampling Rates Selection

Abstract

Falls are dangerous for the elderly, often causing serious injuries especially when the fallen person stays on the ground for a long time without assistance. This paper extends our previous work on the development of a Fall Detection System (FDS) using an inertial measurement unit worn at the waist. Data come from SisFall, a publicly available dataset containing records of Activities of Daily Living and falls. We first applied a preprocessing and a feature extraction stage before using five Machine Learning algorithms, allowing us to compare them. Ensemble learning algorithms such as Random Forest and Gradient Boosting have the best performance, with a Sensitivity and Specificity both close to 99%. Our contribution is: a multi-class classification approach for fall detection combined with a study of the effect of the sensors’ sampling rate on the performance of the FDS. Our multi-class classification approach splits the fall into three phases: pre-fall, impact, post-fall. The extension to a multi-class problem is not trivial and we present a well-performing solution. We experimented sampling rates between 1 and 200 Hz. The results show that, while high sampling rates tend to improve performance, a sampling rate of 50 Hz is generally sufficient for an accurate detection.

Influence of the Antenna Orientation on WiFi-Based Fall Detection Systems

“…For this reason, it is necessary to determine the rate at which the falling event can be registered and select the sensor that best suits these conditions. However, the placement of the sensor on the human body directly affects the detection performance of the system [ 5 , 17 , 18 ]. Additionally, wearable sensors can cause discomfort for users who are not accustomed to wearing such devices.…”

Event-Centered Data Segmentation in Accelerometer-Based Fall Detection Algorithms

“…Hsieh et al [ 24 ] presented an adaptive approach where the duration of the impact window depended on the amplitude of the largest acceleration peak on record. In [ 25 ], an impact window of 1.5 s before and a window of 0.5 s after the largest acceleration magnitude was taken. Two additional windows were then placed before and after the impact window.…”

“…The same as in [ 18 ], researchers in [ 20 , 21 , 22 ] also employed a single window for event-centered data segmentation. Another approach was used in [ 19 , 23 , 24 , 25 , 26 ], where data segmentation was based on multiple windows. Using more than one segmentation window is justified by an idea that in spite of the variable and irregular nature and typology of falls, they can be decomposed as a sequence of typical “stages” or phases.…”

A Hierarchical Approach to Activity Recognition and Fall Detection Using Wavelets and Adaptive Pooling

“…In this work, we make use of the SisFall dataset to perform fall detection with direction and severity and activity of daily living detection since it has been the dataset of choice in multiple works addressing the fall detection domain [ 37 , 38 , 39 ] as it includes recordings of volunteers from various ages (ages from 19 to 75 years), has diversity in the gender make up of the participants (19 males and 19 females from a total of 38 volunteers) and is one of the biggest datasets available in terms of the type of falls and activities being recorded. Since both accelerometers are placed at the same position and therefore measure the same movements, data from only one of the accelerometers along with the gyroscope are considered in this work.…”