FALL DETECTION USING BIOLOGIALLY INSPIRED

MONITORING

Artificial Immune System in the Area of Assisted Living

Sebastian D. Bersch, Djamel Azzi

and Rinat Khusainov

School of Engineering, University of Portsmouth, Portsmouth, U.K.

Keywords: Artificial Immune System, AIS, Fall Detection, Health Monitoring, Wellbeing, Accelerometer.

Abstract: This position paper supports the use of Artificial Immune System (AIS) in the area of Ambient Assisted

Living (AAL). While AIS has been used for anomaly detection and classification in a wide range of

applications, little work has been done on using AIS for detecting abnormal behaviour in health monitoring

applications. In this paper, we propose to use AIS for fall detection, since falls can be seen as deviations

from the normal behaviour. We justify our proposal by analysing research that has been carried out in the

past using AIS in different fields and emphasising on the similarities to the area of AAL. The paper also

describes the experimental setup that is currently being used for our current and future work.

1 INTRODUCTION

Due to the advances in medicine over the last half

century, humans are able to live longer than ever

before. Concurrently the birth rate is slowing down.

These trends invert the aging pyramid (Figure 1),

meaning that soon there will be more people over

the age of 65 than under (Commission et al. , 2009).

In economical terms: fewer caretakers will have to

look after more elderly people. The field of Ambient

Assisted Living (AAL) aims to find solutions for

this problem.

Figure 1: Aging pyramid inversion.

The research directions in AAL are widely varied,

but the main aim is common - to support and help

the elderly to stay longer safe and healthy in their

home in a cost effective manner. The work ranges

from sensor development for better vital sign

monitoring (Boylan, 2011) and position information

to recognition of Activities of the Daily Living

(ADL) (Sim et al., 2010) over to health monitoring

(Monekosso and Remagnino, 2010) (see Figure 2).

Each monitoring system will learn at one stage what

“normal” health means. The problem is that the

definition of “normal” is slightly different for each

person and therefore introduces uncertainty in the

results of each system monitoring the behaviour of

an elderly person. So far researchers have

concentrated on short term behaviour monitoring.

These research projects included the use of Hidden

Markov Model (HMM) (Monekosso and

Remagnino, 2010) and Rule Based Activity

Recognition (Storf, Becker, & Riedl, 2009). A new

direction is to move away from the short term

monitoring and consider the long term monitoring of

a person (Elbert et al., 2011).

Figure 2: Range of monitoring.

This long term monitoring cannot only be used

for behaviour analysis but also to improve fall

detection. Previously, we investigated intelligent fall

detection for elderly people (ADL in Figure 2) using

Fast Fourier-Transformation (Bersch et al., 2011)

and neural networks to successfully detect falls. This

320

D. Bersch S., Azzi D. and Khusainov R..

FALL DETECTION USING BIOLOGIALLY INSPIRED MONITORING - Artiﬁcial Immune System in the Area of Assisted Living.

DOI: 10.5220/0003674803200323

In Proceedings of the International Conference on Evolutionary Computation Theory and Applications (ECTA-2011), pages 320-323

ISBN: 978-989-8425-83-6

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

position paper makes a case to use Artificial

Immune System as a new way for adaptive long

term monitoring.

2 ARTIFICIAL IMMUNE

SYSTEMS AND ASSISTED

LIVING

Artificial Immune Systems – AIS is a fairly new

approach in Artificial Intelligence and is based on

modelling the human immune system. The technique

is very similar to genetic algorithms (GA) in terms

of binary detectors used to match data strings with

the main difference that it is better suited to detect

system anomalies and abnormal behaviour patters

(Jui-Yu Wu;, 2010).

For a better understanding of the AIS function,

the human immune system needs to be briefly

described. The general function of the immune

system (IS) is to detect cells in the human body and

class them using the chemical surface structure

(matching) into a “non-self” and a “self” set. While

the “self” set is harmless to the body and is a

repetitive pattern in the body, the “non-self” set is

harmful and only a sporadic pattern in the body,

besides a chronic illness. This is in technical terms a

system anomaly and called a pathogen in IS terms.

When the human IS detects a pathogen, it has to be

attacked and destroyed. This can be illustrated by

looking at the different illness symptoms people

have with different infections. (Timmis, 2007)

(Hofmeyr, 2000).

Case for AIS in Assisted Living

– In the work of

(Hofmeyr, 2000), the authors point out that for AIS

to be effective, an application should have the

following features:

• Require pattern classification and response

• Require a distributed architecture scalable to

environments with arbitrary numbers of nodes

• Address problems for which there is some

commonality of patterns across the nodes, i.e.,

multiple nodes see the same or similar patterns

within some limited time period

• Require the detection of novel anomalous

patterns

• Change behaviour slowly over time

• Have storage capacities on any single node that

are small compared to the amount of

information required to represent all possible

normal patterns

These application requirements need to be compared

to the future research areas listed in the Ambient

Assisted Living Roadmap of the AALIANCE

(Broek et al., 2009) to evaluate the opportunities of

AIS in AAL. The key research areas are:

• Detection of emergency situations

• Activity recognition

• Handling imperfect information

• Fusing sensor data

Analysing the AAL research areas, the main

qualities that are needed are data fusion and pattern

recognition, which matches the ones for an effective

AIS implementation.

Previous Work – AIS has been used in a wide range

of research areas including fault detection and

monitoring systems. The past research in AIS has

mostly looked into detection of deviations from

normal behaviour.

In the research work of Cai et al. (Yixin Cai et

al., 2010) the authors are comparing Artificial

Immune Recognition Systems (AIRS) versus

Artificial Neural Networks (ANN), Logistic

Regression (LR), Support Vector Machines (SVM)

and K-Nearest Neighbour (KNN) for their ability for

fault detection. The authors used different machine

learning algorithms (including AIRS) to classify real

power distribution fault data from three regions in

North Carolina. AIRS outperformed the other

algorithm most of the time.

Polat et al. (Polat, Expert Systems with

Applications, 2008) used a combination of principle

component analysis (PCA) and AIRS to classify the

UCI lung cancer data set. The authors were able to

achieve 100% accuracy on the data set and pointed

out that this accuracy is the highest among the

classifier reports in the literature. The same authors

also used a similar combination of AIRS with PCA

in (Polat, Expert Systems with Applications, 2008)

to classify EEG signals. Nearly the same accuracy

was achieved. The authors point out that the most

important feature of AIRS is the ability of its self-

learning and therefore enable a fully automated

classification.

Another application of AIS is the area of

Intrusion Detection Systems (IDS). The authors of

(Golovko et al., 2010) proposed to use AIS in IDS to

detect unknown intrusion attacks. Furthermore,

Hofmeyr et al (Hofmeyr, 2000) designed an

architecture for an AIS and developed a network

IDS called LISYS as a proof of concept. LISYS

used the simulated data as a “self” set and was then

confronted with seven different unknown attacks to

the system. The outcome of the test was that all

seven intrusive activities were detected. A low false

positive rate was achieved without compromising

the ability to detect intrusions.

FALL DETECTION USING BIOLOGIALLY INSPIRED MONITORING - Artificial Immune System in the Area of

Assisted Living

321

In the field of monitoring systems Lehmann et.al

(Bersini and Carneiro, 2006, pp. 335-348) used AIS

to control the heating system of an intelligent home.

The aim was to learn the inhabitants’ behaviour and

adjust the heating pattern for the house accordingly.

AIS was used to act quickly on the users’ special

heating demands without forgetting its normal

behaviour. The AIS was able to adapt and could still

react to changes in the users behaviour.

Considering the long term monitoring of the

wellbeing of elderly people, several comparisons can

be drawn to past work in AIS. Using the circadian

activity rhythms (CAR), proposed in (Virone, 2008),

as the foundation of the rhythm of an elderly person

it is possible to speak of a normal repetitive pattern

in living conditions, which is a necessity for AIS.

The probability of a health related problem with the

monitored person increases, if a deviation of its

“normal” pattern occurs (Monekosso and

Remagnino, 2010) (Storf et al., 2009).

In the work of Franco et al. (Franco et al., 2010),

the nycthemeral shift (change in daily routines) is

used to detect dementia related diseases. This shift

represents an abnormality in a recurring pattern.

Parallels can be drawn to IDS. Intrusion Detection

Systems based on AIS use frequent events to build a

“self” set and identify attacks through the deviation

from “normal” traffic pattern (compare (Golovko et

al., 2010) and (Hofmeyr, 2000)).

The research described above supports the

believe of the authors that AIS will also achieve

good results in the area of fall detection and in

particular the feature of self-determination of AIS

which should help improve the accuracy of fall

detection and reduce false alarms in ubiquitous fall

detectors.

3 EXPERIMENAL SETUP

Equipment and Data Collection – The accelerometer

data that will be used for the proposed research was

collected in an earlier experiment. The hardware

used to collect the data was an inexpensive off-the-

shelf development kit. Two Texas Instruments

eZ430-Chronos 868 MHz sports watches were used

with an optimised firmware (customised to achieve

equal time spacing between the accelerometer

samples). One watch was attached to the subject’s

waist, the other to its wrist. The watch samples the

X, Y and Z axes each 1/10

of a second, and sends

the data to a PC. Compared to other research

experiments, which use higher sampling rates (up to

160 Hz) (Cagnoni et al., 2009), a sampling rate of 10

Hz can be considered as quite low. The use of a

lower sampling rate has three main benefits:

1. Lower data generation

2. Longer battery life of the wearable sensor

3. Short peak accelerations are ignored

The Data Set – The pre-recorded data collection

included three different activities, walking, sitting

and falling. The records of each activity were 2

minutes long and up to 10 repetitions were available.

As a consequence of the research looking into long

term health monitoring, this data needed to be

enlarged. The different data snippets of each activity

were randomly linked together to render a simulated

month worth of data. The requirements on the newly

created data sets were:

• Sitting had to be a continuous activity

(occurring every 4 hours)

• Falling had to be a sporadic event (once

every 10 days).

The longer time frame of the data set allows the

AIS to be trained with different walking and sitting

behaviours of one test person. This will help to build

a “normal” behaviour pattern and a fall should be

detected more easily because of the deviation from

normal activity.

Preparation of Data and Implementation of AIS –

The first implementation of the monitoring system is

designed to validate the use of AIS in the field of

AAL applications. The first stage should

demonstrate that the particular used data set contains

a normal as well as an abnormal self set. A data

instance presented to the AIS contains four

attributes:

1. X-Acceleration

2. Y-Acceleration

3. Z-Acceleration

4. Acceleration Average (calculated using

equation (1)

 =







−







−







−





(1)

In this phase, only the data set of the arm is used and

no pre-processing takes place, besides a simple

averaging over the last 5 data samples. Each data

instances can be presented as a 32 Bit variable.

Detectors are matched with the data instances using

the Hamming Distance to calculate the bit difference

between both instances (see Figure 3).

Figure 3: Presentation of a data instance and matching.

ECTA 2011 - International Conference on Evolutionary Computation Theory and Applications

322

4 FUTURE WORK

The next stage is the implementation of a monitoring

system based on the described AIS method. The

direct outcomes are to achieve a proof of a “non-

self” and “self” set in the used acceleration data set

and a dynamic adaptation in immature and mature

detectors. A dynamic variation in the monitoring can

be achieved by mutating detectors at specific time

intervals (lifespan). The mutation process can be

based on different method like random generation or

more advanced clonal algorithms. Therefore further

research will look into the following areas:

• Possible use of data pre-processing

• Generation of detector

• Lifespan of detector

• Matching of detector and data instance

• Uncertainty after a mature detector is

triggered

The authors believe that research in these areas will

lead to an improved recognition in fall detection.

The research outcome will be compared against the

earlier results using FFT and neural networks.

ACKNOWLEDGEMENTS

This work is supported by the University of

Portsmouth under the Higher Education Innovation

Fund (HEIF 4) and performed by the Digital

Wellbeing Research Group at the same University.

REFERENCES

Yixin Cai, Mo-Yuen Chow, Wenbin Lu, & Lexin Li;.

(2010). Evaluation of distribution fault diagnosis

algorithms using ROC curves. Power and Energy

Society General Meeting, 2010 IEEE , 1 - 6.

Virone. (2008). Monitoring activity patterns and trends of

older adults. Engineering in Medicine and Biology

Society, 2008. EMBS 2008. 30th Annual International

Conference of the IEEE , 2071 - 2074.

Bersch, S., Chislett, C., Azzi, D., & Khusainov, R. (2011,

Jan 1). Activity detection using frequency analysis and

off-the-shelf devices: fall detection from accelero-

meter data. eprints.port.ac.uk .

Bersini, H., & Carneiro, J. (2006). Artificial Immune

Systems: 5th International Conference, ICARIS 2006,

Oeiras, Portugal, September 4-6, 2006, Proceedings

(Lecture Notes in Computer Science).

Boylan, G. (2011, Jan 1). EEG monitoring in the neonatal

intensive care unit: A critical juncture. Clinical

neurophysiology: official journal of the … .

Broek, G., Cavallo, F., Odetti, L., & Wehrmann, C. (2009,

Sep 22). Ambient Assisted Living Roadmap. 1-120.

Cagnoni, S., Matrella, G., Mordonini, M., Sassi, F., &

Ascari, L. (2009). Sensor Fusion-Oriented Fall

Detection for Assistive Technologies Applications.

Intelligent Systems Design and Applications, 2009.

ISDA '09. Ninth International Conference on , 673 -

678.

Commission, E., Economic, D.-G., & Affairs, F. (2009).

Ageing Report 2009.

Elbert, D., Storf, H., Eisenbarth, M., Ünalan, Ö., &

Schmitt, M. (2011). An approach for detecting

deviations in daily routine for long-term behavior

analysis. Workshop: Orange Alerts - Behaviour

Modeling and Health of older people in their homes

(AAL 2011). Dublin.

Franco, C., Demongeot, J., Villemazet, C., & Vuillerme,

N. (2010). Behavioral Telemonitoring of the Elderly at

Home: Detection of Nycthemeral Rhythms Drifts from

Location Data. Advanced Information Networking and

Applications Workshops (WAINA), 2010 IEEE 24th

International Conference on , 759 - 766.

Golovko, V., Komar, M., & Sachenko, A. (2010).

Principles of neural network artificial immune system

design to detect attacks on computers. Modern

Problems of Radio Engineering, Telecommunications

and Computer Science (TCSET), 2010 International

Conference on , 237 - 237.

Hofmeyr, S. (2000, Jan 1). Architecture for an artificial

immune system. Evolutionary Computation .

Jui-Yu Wu;. (2010). Computational Intelligence-Based

Intelligent Business Intelligence System: Concept and

Framework. Computer and Network Technology

(ICCNT), 2010 Second International Conference on ,

334 - 338.

Monekosso, D., & Remagnino, P. (2010). Behavior

Analysis for Assisted Living. Automation Science and

Engineering … .

Polat, K. (2008, Jan 1). Artificial immune recognition

system with fuzzy resource allocation mechanism

classifier, principal component analysis and FFT

method based new hybrid automated …. Expert

Systems with Applications .

Polat, K. (2008, Jan 1). Computer aided medical diagnosis

system based on principal component analysis and

artificial immune recognition system classifier

algorithm. Expert Systems with Applications .

Sim, K., Yap, G.-E., Phua, C., Biswas, J., Phyo Wai, A.,

Tolstikov, A., et al. (2010). Improving the accuracy of

erroneous-plan recognition system for Activities of

Daily Living. e-Health Networking Applications and

Services (Healthcom), 2010 12th IEEE International

Conference on , 28 - 35.

Storf, H., Becker, M., & Riedl, M. (2009). Rule-based

activity recognition framework: Challenges, technique

and learning. Pervasive Computing Technologies for

Healthcare, 2009. PervasiveHealth 2009. 3rd Interna-

tional Conference on, 1 - 7.

Timmis, J. (2007, Jan 1). Artificial immune systems—

today and tomorrow. Natural Computing.

FALL DETECTION USING BIOLOGIALLY INSPIRED MONITORING - Artificial Immune System in the Area of

Assisted Living

323