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.

Falls are extremely damaging to the elderly. The number of elderly who have experienced falls has increased over the years, several of the elderly stay alone or in in badly maintained elderly homes. This makes a low-cost fall detection system a necessity. There has been huge improvements in terms of IoT systems, ML algorithms. Varied data sets have been collected across the world for fall detection. These data sets have a very little in common among them, in terms of user demographics, sensors used, the ADL and Fall activities Hence in this paper we present a data set that has wide user demographics, we used various sensors – such as accelerometer, gyroscope, magnetometer and hear rate. We used wrist worn sensors to collect data. In this paper we present a detailed analysis of the data set we collected using common ML algorithms such as – Naïve Bayes (NB), K Nearest Neighbor (KNN), Logistic Regression (LR), Random Forest (RF) and Support Vector Machines (SVM). We analyzed the performance of these algorithms for variations in accuracy with respect to age, gender, height, weight and health issues and we have identified outliers by analyzing each incorrect prediction. This paper provides the complete details of the data collection methodology, The methods used for analysis and presents the results of analysis in complete detail.

Fall of an elderly person often leads to serious injuries and death. Many falls occur in the home environment, and hence a reliable fall detection system that can raise alarms with minimum latency is a necessity. Wrist-worn accelerometer-based fall detection systems and multiple datasets are available, but no attempt has been made to analyze the accuracy and precision. Wherever the comparison does exist, it has been run on a cloud. No analysis of the models, convergence, and dataset analysis on Systems on a Chip (SoCs) has ever been attempted. In this paper, we attempt to present why Machine Learning (ML) algorithms in their current state cannot be run on existing SoCs. We have used Snapdragon 410c SoC to do our analytics. In this paper, we have used the k^th-nearest neighbour to prove that ML cannot be directly run on SoCs. We have looked at the effect of distance metrics and neighbors as well as the effect of feature extraction on the accuracies and the latencies. In this paper, we establish the need for model compression and data pruning for fall detection using ML/Deep Learning algorithms on SoCs. We have done this by analyzing various datasets on varying architectural parameters.

Falls, especially those left unattended, are fatal for the elderly. Several efforts have been made to use the Internet of Things and Machine Learning algorithms to detect falls. All such systems have issues such as (a) sensor placement on the torso and thigh which will be uncomfortable for the elderly. (b) Predictions made on the cloud- the result of the prediction is then sent back to the end device to raise an alert. This is prone to network connectivity and latency issues. We have built an end/device that is wrist-worn and has multiple Inertial Measurement Unit sensors and a heart sensor. We have developed three novel ensemble algorithms (a)Stack(A) (b) Variable weighted Ensemble Voting algorithm-B(VWE(B)), and (c) Variable weighted Ensemble Voting algorithm-C (VWE(C)). Since the ensemble algorithm is run on the end-device built around Qualcomm Snapdragons 820c, we do both feature extraction and selection to reduce data dimensionality. We have used multiple methods such as (a) Identifying features that have maximum impact (b) Principal Component Analysis (PCA) (c) Shapley’s values (d) Cross-Correlation combined with relliefF. We got an accuracy of 97% and specificity of 99%. In this paper, we present the analysis of the system with and without pruning.