Inertial measurement and heart-rate sensor-based dataset for geriatric fall detection using custom built wrist-worn device

This paper describes a dataset acquired from 41 volunteers performing 16 Activities of daily livings (ADLs) and 8 Falls repeated 5 times. This data was collected using a custom wrist-worn end device. The dataset has data collected from Inertial measurement unit (IMU) and heart-rate sensors. The end device is built using Qualcomm Snapdragon 820c System on Chip (SoC) interfaced to the sensors via Interconnect Integrated Circuit (I2C) protocol. The data was sampled for every activity at a rate of 20 Hz for the motion sensors and at a rate of 1 Hz for the heart-rate sensor. The motion sensor comprised of a triaxial accelerometer, triaxial gyroscope, triaxial magnetometer and a linear accelerometer. The heart-rate sensor was medical grade and all sensors were calibrated for the wrist -worn position. The dataset is available on this website https://shamanx86.github.io/fall_detection_data/ and https://doi.org/10.5281/zenodo.10013090.


Subject
Computer Science in Healthcare Specific subject area Geriatric Fall detection Data format Raw data of triaxial accelerometer, triaxial gyroscope, triaxial magnetometer, linear accelerometer and Heart-rate sensor with time stamp.Type of data Table, each table having six columns, time-stamp, x axis data, y-axis data and z axis data, number of axis and type of sensor(label) except in case of heart-rate where there will be only three columns, time-stamp, beats per minute and label (hrt for heart-rate) Data collection Data was collected using a custom built-wrist worn end worn on the left-wrist.Qualcomm Snapdragon 820c.We used MAX30102 Heart rate and SP02 sensor, MPU6500, which gives 3-axis acceleration, 3-axis linear acceleration and 3-axis gyroscope data and GY273 Magnetometer chip for data collection.All the sensors are interfaced to the SoC via the I2C interface using a Mezzanine board.

Value of the Data
• This is the only dataset available that uses only wrist worn data and has a total of 4920 instances.• Though there are other wearable based datasets available, the sensors are worn either as a combination of the thigh, waist, torso, ankle or all of the above.• We have thoroughly analysed the performance on various ML algorithms and obtained accuracies and precision in detecting Falls as high as 97%.• Highly scalable as the device used to collect the data is a low-cost device that can be easily worn by any geriatric unlike image or acoustic based dataset.

Data Description
The data-set and the code is available in [ 1,2 ].The steps to download the data is shown in Fig. 1 .
The actual map to download the .csvfrom the website is illustrated below in Fig. 2 .Also the entire dataset in a zip file is available at: https://doi.org/10.5281/zenodo.10013090

Format of the csv file
Fig. 3 gives the snapshot of the .csvfile.
Column A gives the time in which the data was recorded.In case of the triaxial accelerometer, gyroscope, magnetometer and linear acceleration.The next three column i.e.B, C and D give x, y and z axis values and column E gives the number of axis and column F gives the name of the sensor in short.In case of heart-rate only four columns are used, column A is the time stamp, column B is the number of heart-beats per minute, column C indicates a single value and column D gives the sensor name in short which works as a label.The expansion is as follows "acc" -3 axis accelerometer gives acceleration along x,y and z plane."acg" -3 axis linear accelerometer give linear along x,y and z plane "gyro" -3 axis gyroscope gives the rotational moment along x,y and z plane "mgm" -3 axis magnetometer gives the position with respect to the earth's gravitational force along x,y and z plane "hrt" -single value of heart-rate in beats per minute

Experimental Design, Materials and Methods
In this data article, the data was collected from a total of 41 volunteers performing 16 ADLs and 8 Falls listed in Table 1 .Every activity was repeated for 5 trials.
The collection of data from various sensors was done using a python code and was perfectly timed using the inbuilt timer of the 820c.Each of the activities produced two Comma separated Variable (.csv) files, one had data from heart-rate sensor and the another which did not have the data from the heart-rate sensor.All Falls were done in the safety of a well-padded anechoic chamber.The volunteer was asked to wear the wrist-worn device on his/her left wrist while performing all the activities.Before starting each activity, the base heart-rate of the volunteer was recorded and it was ensured that they were at their base heart-rate when they started the activity.The placement of the sensor is shown in Fig. 3 below along with the internal circuitry of the wrist worn device.This system is built around a powerful System on Chip (SoC) that is Qualcomm Snapdragon 820c [ 3 ].The 820c chip has been developed specifically for wearable and IoT applications.We used MAX30102 [ 4 ] Heart rate and SP02 sensor, MPU6500 [ 5 ], which gives 3-axis acceleration, 3-axis linear acceleration and 3-axis gyroscope data and GY273 [ 6 ] Magnetometer chip for data collection.All the sensors are interfaced to the SoC via the I2C interface using a Mezzanine board.The 820c chip also has inbuilt 802.11 transceiver to send the data wirelessly to the cloud.Also, the SD card slot can be used to store data for long durations (a week or more).The Fig. 4 gives the experimental setup of the wearable device.
Volunteer statistics:  The 41 volunteers were drawn from a wide demographics and it was ensured that they varied in terms of Age, Height, Weight, Gender and pre-existing health conditions.The volunteer statistics is represented using a series of pie chart below in Fig.  .The complete details of the volunteers is given in Table 2 .

Limitations
Not applicable.

Ethics Statement
As our work involved human subjects, an informed consent form as per our university guidelines was taken from every volunteer and witnessed by two people in a standard format provided by the university guidelines.All Female volunteer were supervised by a female faculty as per university guideline.Ethical approval of surveys were not required, only the institute's rules and regulation needed to be followed which we have.

Fig. 1 .
Fig. 1.Steps to access the individual data per activity from the dataset.

Fig. 4 .
Fig. 4. Device used for collecting data and its internal circuit.

Table 1
Activity list and details.

Table 2
Volunteer details.