Wandering Analysis with Mobile Phones

On the Relation Between Randomness and Wandering

Agusti Solanas

, Edgar Batista

, Frederic Borras

, Antoni Mart

ınez-Ballest

and Constantinos Patsakis

Smart Health Research Group, Dept. of Computer Engineering and Mathematics,

Universitat Rovira i Virgili, Catalonia, Spain

Department of Informatics, University of Piraeus, Piraeus, Greece

Keywords:

Behaviour Analysis, Wandering, Randomness, Elderly.

Abstract:

Population pyramids have rapidly changed their shape worldwide. Specially in industrialised countries, where

a steady ageing process is taking place. Due to the ageing of the population, age-related illnesses such as

mild cognitive impairment (MCI) and dementia are becoming commonplace and healthcare systems struggle

to improve the quality of life of patients and carers. Wandering and disorientation are common symptoms

amongst patients with MCI, and they could lead to fatal outcomes. In this article we propose a wandering

detection technique based on the analysis of randomness in patients’ trajectories, which are gathered by means

of standard low-cost mobile phones with GPS.

1 INTRODUCTION

The average age of the world population has in-

creased progressively over the last 50 years as a re-

sult of the reduction of fertility and an increase in

life expectancy. The ageing of the population poses

a formidable challenge for public healthcare systems

since they have to face the rise of an aged and very de-

manding population and their associated health condi-

tions, namely chronic illnesses, injuries and disabili-

ties.

This demographic change results in a huge impact

on society and proper countermeasures have to be put

in place so as to cope with it in the years to come.

The aforementioned ageing of the population leads

to an increase in the cases of cognitive disorders like

Mild Cognitive Impairment (MCI), Parkinson’s dis-

ease (PD) and Alzheimer’s disease (AD), to name the

most common.

MCI is specially interesting because it can be un-

derstood as a precursor of early stages of AD and

PD and other types of early dementia that imply im-

paired memory function whilst the cognitive function

is generally preserved (Petersen et al., 2001). MCI

is a brain function syndrome involving the onset and

evolution of cognitive impairments beyond those ex-

pected based on the age and education of the indi-

vidual, but which are not signiﬁcant enough to inter-

fere with their daily activities (Petersen et al., 1999).

Annual prevalence estimates for MCI in the United

States range from 3% to 4% in the eighth decade

in the general population. Amongst community-

dwelling African Americans, the estimated preva-

lence is 19.2% for those aged 65 − 74 years, 27.6%

for those aged 75 − 84 years, and 38% for those aged

85 years and older. The prevalence of mild cogni-

tive impairment increases with age. The prevalence

is 10% in those aged 70 − 79 years and 25% in those

aged 80 −89 years.

People suffering from MCI and early stages of dif-

ferent types of dementia might experience a decrease

in their cognitive capabilities that might affect their

mobility patterns but they still have considerably high

degrees of autonomy (i.e. they can live alone, walk,

exercise). The most apparent impairment is related

to their memory function: patients might become

spatially and temporally disoriented, and might have

problems in ﬁnding their way home, or they could

forget to accomplish tasks. As a result of the mem-

ory function impairment, patients suffering from MCI

might follow strange trajectory patterns and wander.

Amongst all symptoms related to MCI, namely

memory loss, disorientation, wandering, anxiety, etc.

Wandering is one of the most frightening for relatives

and carers. People that wander might easily get lost

and, as a result, they could put their lives in jeop-

168

Solanas A., Batista E., Borras F., Martínez-Ballesté A. and Patsakis C..

Wandering Analysis with Mobile Phones - On the Relation Between Randomness and Wandering.

DOI: 10.5220/0005329601680173

In Proceedings of the 5th International Conference on Pervasive and Embedded Computing and Communication Systems (PECCS-2015), pages

168-173

ISBN: 978-989-758-084-0

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

ardy. In this article, we focus on detecting wander-

ing/disorientation patterns in trajectories gathered by

GPS-enabled mobile phones.

The rest of the article is organised as follows:

Next, in Section 2 we provide some basic background

and an overview of related work. In Section 3 we

introduce our method to detect wandering and disori-

entation and show that there exists a relation between

the amount of randomness in trajectories and the ap-

pearance of abnormal patterns. Next, in Section 4 we

describe our initial experiments with real patients and

provide an illustrative example with real data. Finally,

the article concludes in Section 5 with some ﬁnal re-

marks and lines for further research.

2 RELATED WORK

Research related to the provision of health services

by means of electronic devices is very wide. With the

aim to provide the reader with a basic background,

in Section 2.1 we elaborate on electronic health (e-

health) and mobile health (m-health), next in Sec-

tion 2.2 we recall the SIMPATIC Project to illustrate a

novel and recent initiative to provide intelligent mon-

itoring of people with MCI.

2.1 e-Health and m-Health

The healthcare sector has adopted the so-called in-

formation and communication technologies (ICT)

and, as a result, the concept of electronic health (e-

health) (Eysenbach, 2001) appeared. e-Health is con-

tributing to the reduction of costs and the increase

of efﬁciency. ICT might be used for a variety of

health-related tasks, namely communication between

patients, doctors and carers, distant provision of care,

remote support to diagnostic, EHR, medication adher-

ence control, etc. e-Health substantially reduces the

displacements of professionals and patients, globally

brings down the cost of medical resources, and makes

treatments and health watchfulness more comfortable

to patients.

After the consolidation of e-health, the general-

ized utilisation of mobile devices with GPS, or other

self-localisation capabilities, (e.g., smartphones) gave

birth to the idea of mobile health (m-health). There is

no doubt about the huge potential of m-Health since

it adds to the advantages of e-health all the beneﬁts

related to the ubiquity of mobile devices (i.e., global

monitoring capabilities, wide availability and imme-

diacy).

Mobile health (m-health) can be deﬁned as the

discipline founded on the use of mobile communi-

cation devices in medicine, or more speciﬁcally, the

delivery of healthcare services via mobile commu-

nication devices, or: “Emerging mobile communi-

cations and network technologies for healthcare sys-

tems” (Istepanian et al., 2006). The use of mobile

devices helps to perform tasks more efﬁciently. Espe-

cially the remote monitoring of patients and the com-

munication between professionals, relatives and pa-

tients will highly beneﬁt from m-health. m-Health

extends the capabilities of indoor monitoring environ-

ments and it is a powerful tool that allows the ad-

vance of several lines of research, namely the con-

tinuous assessment of the state of patients, the early

detection of emergency situations, the detection of

changes in health conditions, the detection of abnor-

mal situations, the early detection of fragile situations,

etc. Some interesting examples of m-health applica-

tions can be found in (Solanas et al., 2013)(Postolache

et al., 2012)(Foundation, 2013).

2.2 The SIMPATIC Project

It is apparent that continuous surveillance can help

people with mild cognitive impairments (MCI) and

dementia but, current methods have several limita-

tions: (i) surveillance could be seen as a serious pri-

vacy invasion (Mart

ınez-Ballest

e et al., 2013), (ii)

monitoring is not autonomous and, (iii) abnormal be-

havioural patterns are neither analysed nor detected.

With the aim to avert some of these limita-

tions, the SIMPATIC project (Mart

ınez-Ballest

e and

Solanas, 2014), has developed a new intelligent and

autonomous system that monitors the location of pa-

tients, who suffer from MCI. Figure 1 shows some

screenshots of the SIMPATIC Project web-based con-

trol environment.

The system behaves autonomously (without the

need for user intervention) and also intelligently (i.e.

it learns from the users and adapts to their be-

haviours), also it is able to detect abnormal users’ be-

haviours. In addition, the system reacts to risky situa-

tions, namely the user is in a dangerous area (e.g. the

edge of a cliff), he/she is not moving for a long time,

he/she is wandering, etc. The SIMPATIC Project,

which is currently ongoing, is collecting location data

from several patients diagnosed with MCI. Specially,

it is focussed on people diagnosed with GDS-3 and

GDS-4

3 WANDERING ANALYSIS

The act of wandering is deﬁned as walk or move in

Global Deterioration Scale (GDS)

WanderingAnalysiswithMobilePhones-OntheRelationBetweenRandomnessandWandering

169

Figure 1: Screenshots of the web-based control interface of

the SIMPATIC Project. From top to bottom: Users regis-

tration page, Trajectory monitoring and ﬁltering page and,

alarms log page.

leisurely or aimless way or, travel aimlessly through

or over an area. This behaviour appears in people with

MCI and, if it is properly detected, it could alleviate

further risks of suffering serious injuries as a result of

the disorientation. Our main aim is to propose an au-

tomatic method that can detect wandering by means

of analysing the mobility patterns of monitored pa-

tients.

Our assumption, that we aim to discuss next, is

that trajectories that contain wandering patterns have

more randomness than trajectories free of wandering

or disorientation. Next, we discuss the rationale be-

hind this assumption and we propose a method to

analyse the degree of randomness of a trajectory by

means of graph cycles analysis.

3.1 Rationale

Impairments in memory function, typical in people

with MCI, could translate into abnormal mobility pat-

terns, for example, patients will follow strange paths

to reach their homes, or they will go to the supermar-

ket more frequently (because they forget to buy what

they need).

These abnormal mobility patterns can be seen as

short-length cycles present in the trajectories of pa-

tients (e.g. disoriented patients going home might

walk around their block until they ﬁnd the right en-

trance). On the contrary, trajectories that are clearly

guided (i.e. without disorientation) will have long cy-

cles (e.g. patients leave their home, go to the cinema,

and go back home).

Thus, our goal is to design a system that allows

us to detect the percentage of short and long cycles in

the trajectories of patients. Then, if our system detects

short cycles it could raise a wandering alarm.

3.2 Wandering and Randomness

With the aim to shed some light over our assumption,

we have simulated several trajectories with a variable

amount of randomness, and then we have analysed

the number and length of the cycles found in those

trajectories.

Figure 2 shows three trajectories that we will use

as examples to illustrate our point. In the ﬁrst trajec-

tory (Figure 2-left) we have deﬁned three locations

and we have simulated a regular, deterministic way

to reach them and go back to the origin. Since we

are considering a regular squared grid, at every given

location, the next move is selected amongst 8 possi-

bilities so that the distance to the destination is min-

imised. In the second trajectory (Figure 2-centre), at

each step of the trajectory, we have a 0.5 probabil-

ity of choosing a random move and 0.5 probability of

moving like in the ﬁrst trajectory. Finally, the third

trajectory is completely random (i.e. at every step of

the trajectory, the next move is chosen uniformly at

random from the possible eight options).

Once the three trajectories are simulated, they are

represented as a graph G = {V, E}. Where V , the ver-

texes, are the center of each cell of the grid and E, the

edges, are the transitions from one cell to the next.

After creating each graph, we apply an algorithm

to determine the number and length of Eulerian cy-

cles. This way, we detect cycles of length

2, 3, ... and

we count how many cycles of each length are found.

This information if plotted in Figure 3.

The lenght is measured as the number of nodes in the

cycle.

PECCS2015-5thInternationalConferenceonPervasiveandEmbeddedComputingandCommunicationSystems

170

Figure 2: Simulated trajectories. From left to right: deterministic trajectory, semideterministic trajectory with 0.5 random

probability and, random trajectory.

Figure 3: Percentages of cycles depending on their length. From left to right: cycles for the deterministic trajectory, cycles

for the semideterministic trajectory and, cycles for the random trajectory.

It can be observed that the ﬁrst trajectory (with 0

probability of randomness) has a single cycle with a

length of 56 steps. Thus, 100% of cycles are long.

In the second case (with 0.5 probability of random-

ness), there is a large number of cycles of short length

(90%), whilst only 10% are long. A similar result

is obtained with the third trajectory (with 1 probabil-

ity of randomness) and 100% of the found cycles are

short.

From these preliminary experiments, it could be

inferred that adding randomness to a guided trajec-

tory leads to an increase in the number of short cycles.

Thus, although it has not been proven with statisti-

cally sound experiments, we aim to use our assump-

tion to detect wandering by analysing the percentage

of short cycles in a graph.

4 INITIAL RESULTS WITH REAL

PATIENTS

With the aim to test our proposal with real patients

(within the SIMPATIC Project), we have collected

the trajectories of people diagnosed with MCI and we

have analysed their behaviour. To do so, we have se-

lected a group of 15 people aged from 65 to 83 di-

agnosed with MCI (GDS-3, GDS-4). Each patient

has been provided with a mobile phone equipped with

GPS and a data plan to send their data to our central

server every three minutes. The procedure illustrated

in Figure 4 works as follows:

First, our server receives raw data from the pa-

tients’ smartphones every three minutes and stores

them in a MySQL database. Each packet of data con-

tains the latitude and longitud provided by the GPS, a

time-stamp, the values of the accelerometers, the state

of the smartphone battery and the temperature. Next,

every 15 minutes the data are converted from the op-

erational MySQL database into a strategic PostGIS

database with a PHP script. Once we have the infor-

mation in the PostGIS database, we study it and we

ﬁnd the number of cycles of the patient’s route with

a Java program and an API library (designed ad-hoc).

After obtaining the results, they are sent for visuali-

sation and further analysis to three different outputs,

namely a web viewer to use OpenStreetMaps, R to

plot the results of cycles and Pajek to plot the result-

ing graphs, G, of the trajectories.

For the sake of clarity we include an example of

the results obtained with a real patient. In Figure 5

(left) we show a trajectory followed by a patient dur-

ing 3 days in the area of Tarragona. Figure 5 (right)

depicts the graph created by our system. This graph is

analysed to determine the number and length of cycles

and the results are shown in Figure 6 (20x20 meters

cells are considered).

WanderingAnalysiswithMobilePhones-OntheRelationBetweenRandomnessandWandering

171

Figure 4: Data ﬂow scheme of our proposal. First the raw data are collected from patients, then they are stored and analysed

in our servers, and ﬁnally the data are visualised in Open Street Maps, R and Pajek.

Figure 5: Trajectory and graph of a real volunteer patient of

the SIMPATIC project pilot test.

5 10 15 20 25

0 2 4 6 8

cycle length

percentage (%)

Figure 6: Results of the cycles analysis of the above trajec-

tory/graph.

It can be observed that the percentage of short cy-

cles (under 7, 8 steps) is very low and most of the

detected cycles have lengths between 12 and 18 cy-

cles. These results indicate that the patient behaved

normally during the analysed days, which was con-

ﬁrmed by directly asking the patient.

At the time of writing this article, we are still col-

lecting data and the pilot test is ongoing. However, to

the date, the system has been able to properly identify

alarms and the initial results for wandering detection

(as stated in this article) seem very promising.

5 CONCLUSIONS AND OPEN

RESEARCH LINES

Wandering is a serious problem for people suffering

from MCI and ﬁrst stages of dementia. In this article

we have proposed the idea of studying the existence

of wandering behaviours by means of analysing the

randomness of the trajectories by counting the num-

ber of Eulerian cycles and their lengths.

Although there are still many techniques to be

studied and compared (Lin et al., 2014), the experi-

mental results support our claim that there is a rela-

tion between the amount of randomness and the pres-

ence of wandering. Also, it has been shown that there

is an initial evidence to think that the percentage of

short-cycles in a graph might be a good indicator of

the randomness of a trajectory.

Currently, the presented solution is used in a pi-

lot test within the context of the SIMPATIC project

with volunteers in the Tarragona area. Although the

project is already providing interesting results and a

good service to patients and carers, there are many

open research lines that will deserve further study in

the years to come and are brieﬂy listed as follows:

• Tuning the System. Although the proposed solu-

tion is operative and is currently being tested with

real patients, there is still a lot of work to do to

improve its resilience. Also, there are a number

PECCS2015-5thInternationalConferenceonPervasiveandEmbeddedComputingandCommunicationSystems

172

of parameters that have to be carefully tuned. It

would be desirable to have an automatic system

that determines the values of these parameters and

adapts them to the needs of each patient.

• The Smart Health Paradigm. The process of

urbanisation is concentrating most of the pop-

ulation in cities. Those cities that are hosting

a very demanding population need to improve

their way of managing resources to guarantee a

proper and sustainable living. As a result, smart

cities have appeared (P

erez-Mart

ınez et al., 2013).

The use of ICT in smart cities aiming at im-

proving a variety of services and providing new

solutions has gained importance. It seems nat-

ural that also health services might be offered

within the context of a smart city: intelligent sys-

tems fed with data collected from sensors, users

and mobile devices, etc, pave the way for the

emergence of new services related to health and

well-being. Hence, the concept of smart health

(s-health) arises (Solanas et al., 2014). Using

the sensing and context-aware infrastructures of

smart cities allows the collection of personalised

data that will help to improve our system.

• Recommender Systems. Using the experience

of other users/patients to predict the behaviour of

new patients is an interesting new approach that

will be used by our system in the near future. Col-

laborative Filtering (Casino et al., 2013b)(Casino

et al., 2013a) systems are good candidates to be

studied and we plan to use them to predict possi-

ble erratic behaviour of patients.

• Psychological Analysis. Although our system is

centred in patients. We are very much interested

in the beneﬁts that it can provide to carers and rel-

atives. At the time of writing this lines, we are

analysing which are the effects of using our sys-

tem on the reduction of anxiety and stress of car-

ers and relatives. Preliminary results indicate that

carers feel less stress because they trust the system

that allows them to locate patients if they get lost

or disoriented. Further studies have to be carried

out to conﬁrm these initial ﬁndings.

ACKNOWLEDGEMENTS

The authors are partly funded by La Caixa through

project “SIMPATIC” RECERCAIXA’12, and by the

Government of Catalonia under grant 2014 SGR 537.

REFERENCES

Casino, F., Domingo-Ferrer, J., Patsakis, C., Puig, D., and

Solanas, A. (2013a). Privacy preserving collaborative

ﬁltering with k-anonymity through microaggregation.

In ICEBE, pages 490–497.

Casino, F., Patsakis, C., Puig, D., and Solanas, A. (2013b).

On privacy preserving collaborative ﬁltering: Current

trends, open problems, and new issues. In ICEBE,

pages 244–249.

Eysenbach, G. (2001). What is e-health? Journal of Medi-

cal Internet Research, 3(2):e20.

Foundation, U. N. (2013). mhealth alliance. Website.

http://www.mhealthalliance.org/about.

Istepanian, R., Laxminarayan, S., and Pattichis, C. S.

(2006). M-Health: Emerging Mobile Health Systems,

chapter Preface. Topics in Biomedical Engineering.

International Book Series. Springer.

Lin, Q., Zhang, D., Chen, L., Ni, H., and Zhou, X. (2014).

Managing elders wandering behavior using sensors-

based solutions: A survey. International Journal of

Gerontology.

Mart

ınez-Ballest

e, A., P

erez-Mart

ınez, P. A., and Solanas,

A. (2013). The pursuit of citizens’ privacy: A privacy-

aware smart city is possible. IEEE Communications

Magazine, 51(6).

Mart

ınez-Ballest

e, A. and Solanas, A. (2014). The SIM-

PATIC project. http://projecte-simpatic.cat.

erez-Mart

ınez, P. A., Mart

ınez-Ballest

e, A., and Solanas,

A. (2013). Privacy in smart cities - a case study of

smart public parking. In PECCS, pages 55–59.

Petersen, R., Doody, R., Kurz, A., Mohs, R., Morris, J.,

Rabins, P., Ritchie, K., Rossor, M., Thal, L., and Win-

blad, B. (2001). Current concepts in mild cognitive

impairment. Archives of Neurology, 58(12):1985 –

1992.

Petersen, R., Smith, G., Waring, S., Ivnik, R., Tangalos,

E., and Kokmen, E. (1999). Mild cognitive impair-

ment: clinical characterization and outcome. Archives

of Neurology, 56(3):303 – 308.

Postolache, G., Gir

ao, P., and Postolache, O. (2012). Perva-

sive and Mobile Sensing and Computing for Health-

care - Technological and Social Issues, chapter Re-

quirements and Barriers to Pervasive Health Adop-

tion. Spriger.

Solanas, A., Martinez-Balleste, A., Perez, P. A., Fernan-

dez, A., and Ramos, J. (2013). m-carer: Privacy-aware

monitoring for people with mild cognitive impairment

and dementia. Selected Areas in Communications,

IEEE Journal on, 31(9):19–27.

Solanas, A., Patsakis, C., Conti, M., Vlachos, I. S., Ramos,

V., Falcone, F., Postolache, O., P

erez-Mart

ınez, P. A.,

Pietro, R. D., Perrea, D. N., and Mart

ınez-Ballest

A. (2014). Smart health: A context-aware health

paradigm within smart cities. IEEE Communications

Magazine, 52(8):74–81.

WanderingAnalysiswithMobilePhones-OntheRelationBetweenRandomnessandWandering

173