Enhancing Elderly Care with

Machine Learning-Based Fall

Detection
ABSTRACT
Health is the major worry, and its impalpability increases with increasing age. Thus, taking
care of elders is a very important responsibility. In such a scenario, technology helps people
by providing living assistance. One of the major causes of health degradation or death in
elders is ‘fall’. In this paper, a fall detection system based on machine learning is proposed.
Now-a-days elderly people live alone at home because of poor conditions, different working
culture of people and due to many other reasons. According to World Health Organization
(WHO) studies, falls cause many accidental deaths. Greatest number of fatal falls is seen in
adults older than 65 leading to health problems or injuries. Thus, elderly people require an
attention at the times of emergencies at their residence because they cannot call for help due
to lack of technology access in rural areas or due to their physical conditions The system
detects falls by classifying different activities into fall and non-fall actions. The dataset
SisFall with a variety of activities of multiple participants is used to calculate features.
Machine learning algorithms such as SVM, Decision trees and Random Forest are used to
detect falls on the basis of calculated features. The system acquires accuracy up to 77% by
using Random Forest algorithm.

1. INTRODUCTION
Due to various reasons such as poor living conditions and changes in working culture, many
elderly people nowadays live alone at home. According to studies conducted by the World
Health Organization (WHO), falls are a leading cause of accidental deaths, with the greatest
number of fatal falls occurring in adults over the age of 65, often resulting in health issues or
injuries [1]. So, elderly people require an attention at the times of emergencies at their
residence because it is hard to call for help due to lack of technology access in rural areas or
due to their physical conditions. Automatic fall detection systems are being implemented to
improve the
 quality of life for elderly people and provide them with living assistance. This is especially
important given that the risk of falling is high among older individuals, those with
Parkinson's disease, and patients in rehabilitation units [3]. Main reasons for occurrence of
falls are physical factors like of muscle weakness, posture, Gait balance, vision, due to old
age, or psychological factors or environmental factors etc. Falls are the major cause of
injuries and hip fractures [4]. Immediate aid is crucial in preventing fatalities resulting from
falls in elderly individuals. To ensure proper treatment and care, fall detection systems play a
critical role.
Fall detection systems have been the subject of various research studies. However, there are
no fixed criteria regarding the sensors used, the calculated features, or the algorithms used to
classify falls. Image processing techniques are used after capturing images from camera to
detect falls. Wearable sensors like accelerometer, gyroscope at knee, wrist, neck, and waist
are used to get the data input. Features are calculated by using sensor readings to get some
meaningful data out of the raw data. Falls also can be detected by using ambient sensors like
IR sensors or movement-based sensors.
One of the major challenges in developing a precise fall detection system is false alarms,
which occur when the system alerts to a fall-like activity but not an actual fall. To address
this issue, many research studies are focused on reducing false alarms and improving the
accuracy of the fall detection system. The primary objective of studies is to detect falls in
daily life activities with high accuracy. Threshold-based algorithms can be used to classify
activities as either falls or non-falls, but if a sensor detects an unusual activity, the probability
of false alarms increases. Machine learning-based approaches are less prone to false alarms
and are therefore favoured for fall detection systems.

This work proposes a real-time fall detection system for monitoring elderly people. The
system utilizes the SisFall open-source dataset, which records gait data using a tri-axial
accelerometer [15]. By using machine learning algorithms, falls are detected after calculating
various features. Three different machine learning algorithms, SVM, decision tree and
Random Forest are implemented and compared for better accuracy and performance.

2. LITERATURE SURVEY
Fall detection systems can be implemented using various methods, such as cameras, wearable
sensors, or ambient sensors. When cameras are used in the system, image processing
algorithms are employed. However, it has been found that 24% of falls are not detected using
this method [3]. Ambient-based sensors, such as motion detectors or passive infrared (PIR)
sensors, can also be used in fall detection systems [7]. However, these sensors are limited to
the particular area in which they are implemented. Wearable sensor-based fall detection
systems are more suitable for elderly people as they can detect falls anytime and anywhere,
unlike vision-based and ambient-based fall detection systems, which are restricted to the
house or particular indoor environments. Additionally, wearable sensors are less expensive
than cameras or PIR sensors. In literature, sensors are used in necklace [8], subjects’ head
band, chest band, waist band, right wrist band, right thigh band, and right ankle band [9],
waist band [10], etc. The pendant connected to mobile phone via Bluetooth and phone
communicates with
 concerned person. The problem with this system is, phone should be in 100 meters range of
pendant; and activities like front bending or front falls are not considered due to false
interpretation of fall [8]. Having too many sensors can lead to missed predictions of accurate
activity. Wearable sensor accelerometer-based techniques are considered better than camera-
based and ambient-based methods.
There are two techniques for fall detection after the collection of data from sensors and
feature calculation. The first technique is threshold-based, where if the reading of the sensor
is above a particular threshold, it can be categorized as a fall. In this technique, the threshold
for each calculated feature is different, and many false alarms may be generated. The second
technique involves machine learning classification, where the calculated features can be
tested on a pre- trained model with high accuracy and a lower probability of false alarms.
For the current work, accelerometer data from a wearable sensor is utilized, which has
already been measured for different activities in the SisFall dataset [15]. The most relevant
features are then calculated, and machine learning models of SVM, decision tree and Random
Forest are trained and tested.
Previous research in the literature has utilized these features for threshold-based fall
detection, while the proposed work employs a machine learning approach on the same
features, resulting in improved accuracy. Additionally, no other paper has utilized the
decision tree algorithm for training and testing this dataset. The current work employs both
SVM, decision tree and Random Forest algorithms and compares their performance.

3. METHODOLOGY

FIG 1: Flow of System

1. DATA COLLECTION
For this project, the SisFall dataset [15] was utilized, which includes data collected from a
waist-mounted ADXL345 accelerometer. The ADXL345 is a tri-axial accelerometer that
measures acceleration values along three axes: x, y, and z, and is energy-efficient, able to
function in low-power mode. The dataset includes acceleration data from six young adults
and one elderly participant, who performed 19 daily life activities and 6 fall activities. For the
project, 11467 samples were used for training, and 4915 samples were used for testing.

FIG 2: Data Collection

2. FEATURE EXTRACTION AND PRE-PROCESSING

There are many features that can be calculated using accelerometer data, and which are
significant. In this work, we have considered the same features SisFall dataset [15]. But they
have used threshold technique and this paper uses machine learning technique to cover all
non- listed activity classification as well. For one acceleration sample i.e., acceleration along
all the three axes, all above features are calculated. For 1 lakh data samples, 1 lakh features
are calculated.
To pre-process the data, MIN-MAX scalar is applied to the sensors data. The purpose of this
filter is to eliminate any noise and unwanted glitches from the data. This filter was chosen
because it has been shown to produce results similar to more complex IIR and FIR filters at

FIG 3: Pre-Processed data

different frequencies, as reported in previous studies. Figure 3 displays the original data and
pre-processed data.
3. Fall Detection – Training and Evaluation
The machine learning algorithms used are Support Vector Machines (SVM), Decision Tree
and Random Forest. SVM is an algorithm that can be used for classification as well as
regression. SVM finds the hyper plane such that it has largest distance from nearest training
data point of any class to provide good separation. For non-linear dataset, it can be possible to
find hyper plane to determine the classification by using kernel function. Data points falling
on each side of the plane belongs to different classes.
To add further detail, decision trees are built recursively by splitting data at each node based
on the feature that provides the most information gain. The information gain is calculated
based on the reduction of impurity in the dataset after the split. The process continues until all
the data points are classified into their respective classes. Decision trees can handle both
categorical and continuous data and can be used for both classification and regression tasks.
However, they are prone to overfitting, and techniques like pruning and ensemble methods
like Random forests are used to overcome this limitation.
So, in summary, the algorithms (SVM, decision tree, Random Forest) are compared based on
their fall detection accuracy. The testing is performed on 4915 data samples, and the most
accurate model is selected for further testing.
[Link]
In case of fall, concerned person should be informed. When real time data is collected and
tested for fall, message or call should be initiated on fall detection.
4. ALGORITHM IMPLEMENTATION

FIG 4: Algorithm Implementation

The Algorithm implementation in Fig 4 is done using python 3.

5. OUTCOMES
To evaluate various trained models and to compute the efficiency of various algorithms,
different parameters such as: confusion matrix, sensitivity, specificity, accuracy, training time
and prediction time are computed.

Table 1: Performance comparison of ML Algorithm

The performance of Random Forest is found to be better than SVM and Decision Tree for the
considered data. SVM has the least accuracy and high training and testing time than the other
algorithms. Decision tree may have the less training and testing time than Random forest but
it
 has less accuracy which leads to the false alarms. The are also limitations to Random Forest
like taking more time for training data.

6. CONCLUSION
This article describes a fall detection system that uses wearable sensors, specifically designed
for the elderly. The system employs machine learning algorithms to detect falls from a range
of daily living activities. Compared to the threshold method, the machine learning technique
is preferred because it results in fewer false alarms due to the use of pre-trained gait patterns.
The Random Forest gives more accuracy than SVM and Decision tree. The models are
evaluated using the parameters such as accuracy, training time, testing time, confusion
matrix. Falls are approximately detected by Random Forest with an accuracy of 77%.
Further improvement in accuracy can be obtained by training the models with large dataset
and by identifying optimal features.

REFERENCES
[1] World Health Organization, World Report on Ageing and Health, World Health
Organization, Geneva, 2015.
[2] WHO Global Report on Falls Prevention in Older Age, World HealthOrganization,
Geneva, 2008. [3] De Backere, F., Ongenae, F., Van den Abeele, F., Nelis, J., Bonte, P.,
Clement, E., Philpott, M., Hoebeke, [Link], S., et al.”Towards a social and context-
aware multi-sensor fall detection and risk assessment platform”, Computers in biology and
medicine 64,307/320, 2015.
[4] Khan, M.S, Yu, M, Feng, P, Wang, L, Chambers,”An unsupervised acoustic fall detection
system using source separation for sound interference suppression”,Signal processing 110,
199/210, 2015. [5] Glen Debard, Greet Baldewijns, ToonGoedem, TinneTuytelaars and Bart
Vanrumste, “Camera-based fall detection using a particle filter Glen”,IEEE, 978/4244/9270,
2015.
[6] Rita Cucchiara,AndreaPrati,RobertoVezzani, "A multi-camera vision system for fall
detection and alarm generation“, Expert Systems, 24, 5, 2007.
[7] AhmetYazar, FatihErden, A. Enis Cetin,” Multi-sensor ambient assisted living system for
fall detection”, ResearchGate Conference Paper ,May 2014
[8] Joseph Santiago, Eric Cotto, Luis G. Jaimesy, IdalidesVergara-Laurens, “Fall Detection
System for the Elderly”, IEEE, 978-1-5090-4228-9/17, 2017.
[10] FalinWu, Hengyang Zhao, Yan Zhao, HaiboZhong,” Development of a Wearable-
Sensor- Based Fall Detection System”, International Journal of Telemedicine and
Applications, Volume 2015, Article ID
576364, 2014.
[11]Diana Yacchirema, JaraSuárez de Puga, Carlos Palaua, Manuel Esteve,”Fall detection
system for elderly people using IoT and Big Data ”, The 9th International Conference on
Ambient Systems, Networks and Technologies , 2018.
[14] Yueng Santiago Delahoz , Miguel Angel Labrador,” Survey on Fall Detection and Fall
Prevention Using Wearable and External Sensors”, Sensors, 14, 19806-19842, 2014.
[15] [Link]
[16] S. Vadivelu, Ganesan, S., Murthy, O. V. R., and Dhall, A., “Thermal imaging based
elderly fall detection”, Lecture Notes in Computer Science, vol. 10118, 541-553, 2017
[17] A. Purushothaman, “Fall detection system using Artificial Neural Network”, in 2nd
International Conference on Inventive Communication and Computational
Technologies(ICICCT 2018), Hotel Arcadia, Coimbatore, 2018.