Multi-criteria Modelling Approach

for Ambient Assisted Coaching of Senior Adults

Martin

Znidar

, Bernard

Zenko

, Alja

z Osojnik

, Marko Bohanec

, Pan

ce Panov

Helena Burger

, Zlatko Matja

and Mojca Debeljak

zef Stefan Institute, Jamova cesta 39, Ljubljana, Slovenia

University Rehabilitation Institute, Linhartova 51, Ljubljana, Slovenia

{helena.burger, zlatko.matjacic, mojca.debeljak}@ir-rs.si

Keywords:

Rule-based systems, Multi-criteria Modelling, Ambient Assisted Living, Elderly Care.

Abstract:

This paper presents and critically discusses an approach for knowledge modelling and reasoning in a system

for monitoring and coaching of senior adults. We present a modular architecture of the system and a detailed

description of the modelling methodology which originates from the ﬁeld of multi-criteria decision modelling

and differs from the commonly used ones in this problem domain. The methodology has several characteristics

that make it ﬁt well to the purpose in this application and initial insights from potential users are positive. A

discussion of the suitability of the proposed methodology for knowledge representation and reasoning in the

given problem domain is provided, with an outline of its potential beneﬁts and drawbacks and a comparison

with the ontological approach.

1 INTRODUCTION

In this paper we describe and discuss an approach for

knowledge modelling and reasoning in a system for

multi-modal long-term monitoring and coaching for

the elderly.

Demographic changes in most of the world are re-

sulting in increased share of older people. An im-

portant aspect of quality of life of this population is

independent living, which has many beneﬁts, but also

presents some risks. This gave rise to various techno-

logical solutions of ambient assisted living (AAL) for

the elderly (Azimi et al., 2017; Li et al., 2015). AAL

solutions usually consist of sensors and communica-

tion infrastructure, data management elements, mod-

elling and reasoning tools and ﬁnally actuators or ser-

vices. There is a need and recent trend to provide

platforms that are capable of managing and exploit-

ing multiple components and services of this kind, to

combine multiple information and knowledge sources

and corresponding reasoning on various levels.

In scope of an international collaborative project,

we are tackling such a task and one of the main chal-

lenges is design of the modelling and reasoning com-

ponents. Design of these components must allow rep-

resentation of relevant data and domain knowledge in

a transparent and understandable manner. At the same

time it must be ﬂexible and general enough to support

updates and extensions of the system’s functionalities

and input/output devices.

Systems of this kind usually employ ontologi-

cal (Chen et al., 2012; Yamada et al., 2007; Fides-

Valero et al., 2008; Herv

as et al., 2013), agent-based

(Ayala et al., 2013), rule-based (Bikakis and Anto-

niou, 2010) or mixed (Rafferty et al., 2017; Bouznad

et al., 2017) approaches for knowledge representation

and reasoning. We have considered these types of

approaches and opted for a rule-based approach that

employs hierarchical multi-criteria models, which are

commonly used in the ﬁeld of multi-criteria decision

support (Greco et al., 2016). These models are trans-

parent and easy to interpret and manage. For each is-

sue that we want to provide monitoring and coaching

for, we use a cascade of three such models, which are

responsible for: (I) situation assessment, (II) coach-

ing action selection and (III) coaching action render-

ing. All the models rely on data transformation com-

ponent for input data preprocessing and criteria con-

struction. The three sequential models together with

the data transformation component represent a modu-

lar pipeline architecture which enables decoupling of

core reasoning components from those that are tied

to the input and output devices. This ensures ease of

maintenance and simpliﬁes addition of new input and

Žnidarši

c, M., Ženko, B., Osojnik, A., Bohanec, M., Panov, P., Burger, H., Matja

c, Z. and Debeljak, M.

Multi-criteria Modelling Approach for Ambient Assisted Coaching of Senior Adults.

DOI: 10.5220/0008066900870093

In Proceedings of the 11th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2019), pages 87-93

ISBN: 978-989-758-382-7

output devices.

We ﬁnd this modelling methodology and archi-

tecture very suitable for the problem at hand, but to

the best of our knowledge, hierarchical multi-criteria

modelling methodology has not been used for such

purposes yet. The methodology as such was used in

AAL domain for decision support purposes, for ex-

ample for quality evaluation of AAL systems (Zavad-

skas et al., 2008; Kara et al., 2017), but not for rea-

soning tasks in an automated system of this kind. In

this paper we offer it for consideration through pre-

sentation of its characteristics and discussion of its

knowledge representation and reasoning features that

are relevant for the tackled problem domain.

2 MONITORING AND

COACHING SYSTEM

The monitoring and coaching system for the elderly,

which is being developed in the scope of the project,

is aimed at continuous monitoring of users in their

home environments, detection of relevant changes in

their activity (for example: abandonment of cooking

meals, disturbed sleeping patterns, potential loneli-

ness etc.) and issuing of corresponding coaching ac-

tions.

2.1 Coaching Actions and Domains

The coaching actions are twofold: they can be issued

to the user directly, or they can be indirect, issued to

the members of the user’s social circle, which then

take part in executing the coaching action. An exam-

ple of a direct coaching action is a context-dependent

list of sleep improvement suggestions that is shown

on the user’s tablet screen. An indirect coaching ac-

tion would be a suggestion to a family member to visit

the user or to engage in an activity together (a walk for

example).

In the scope of the project we are addressing four

domains of monitoring and coaching, which were

identiﬁed as important by the target population: (I)

Mobility, (II) Activity, (III) Sleep quality and (IV) So-

cial activity. For each of these domains we are im-

plementing a continuous monitoring and coaching so-

lution for one or more domain-speciﬁc phenomena.

While the data processing, knowledge modelling and

reasoning infrastructure is shared, each coaching so-

lution relies on a speciﬁc set of sensoring devices and

a speciﬁc set of coaching actions.

There is, however, a long and non-trivial informa-

tion path from sensor readings and domain knowledge

to the ﬁnal high-level coaching action suggestions. It

is described in the following subsection.

2.2 Coaching Pipeline

Each phenomenon that we intend to monitor and

coach about demands a speciﬁc set of input informa-

tion about its context and dedicated reasoning pro-

cesses for automated triggering of actions. We denote

one set of such elements that pertains to one coaching

action as a coaching pipeline (see sketch in Figure 1).

The data-ﬂow starts at the sensors that continu-

ously send raw data to a database. In predeﬁned time

intervals, the coaching pipeline gets activated by an

execution process on the server. This triggers the cri-

teria estimation module to collect the relevant data

from the database and calculate or estimate the values

of all the criteria that are needed as inputs by any of

the three models in the pipeline (the situation model,

the coaching action model, and the coaching action

rendering model).

There are three kinds of transformations that the

criteria estimation module conducts in order to trans-

form raw data into criteria for the models:

• Passing, without processing. There is actually no

transformation in this simplest case. An exam-

ple of it would be passing of the temperature as

sensed by the environmental sensor directly to the

models.

• Processing of data with simple explicit functions

(use of ﬁltering, equations, etc.). An example of

this is heart rate calculation from the ECG signal

of a wearable ECG device or a simple noise or

no noise signal from a thresholded amplitude of

an audio signal.

• Provision of output by a machine-learned model.

An example of this kind is transformation of ac-

celerometer data into activity, such as walking, sit-

ting, etc. This is an example for the case of super-

vised models (in which we have the learning data

about the target class). In unsupervised case, the

machine-learned models can be used to transform

speciﬁc data into criteria such as usual or unusual

activity or situation.

The situation model is the ﬁrst one that is exe-

cuted. It conducts a focused situation assessment, the

one which is relevant to the pipeline’s target phenom-

ena. Next, the coaching action model is used to select

a suitable coaching action based on the outcome of

the situation assessment and other relevant external

factors that can inﬂuence the choice of action, such

as the user’s personal proﬁle information that is pro-

vided by the criteria estimation module (forward ar-

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

Figure 1: Coaching pipeline.

row in Figure 1) or the history and sucess of previ-

ous actions (backward arrow in Figure 1). Of course,

the possible coaching actions are individualized and

also strictly limited to only those accepted by the user.

The selected action is then used as the main input in

the action rendering model, which selects a suitable

modality of presentation of the coaching action (the

way it is presented to the user) based on the selected

action and the context that is relevant for rendering,

such as previous effectivness of various modalities

in similar contexts. The coaching suggestion is then

rendered through a corresponding coaching interface.

The coaching interface is in most cases realised by

an electronic device. In addition, in this project the

coaching actions are also realised through people who

are supporting the user and whom we denote as social

circles or secondary users. In some sense, the inter-

face is extended by the members of the user’s social

circles.

For example: an outcome of a situation model

from the coaching pipeline that is dedicated to social

activity might indicate a decrease in social activity in

the past week. The coaching action model would take

this information, together with some additional con-

textual information (as deﬁned in the coaching ac-

tion model) and would suggest a suitable coaching

action, which might be: Go for a visit. However, this

coaching action could be rendered or conveyed to the

user via several modalities, such as a notiﬁcation on

the smartphone screen, speciﬁc audio, visual or tac-

tile nudge, or it might be sent to a secondary user (a

friend), who could convey this message to the user or

accompany the user to a visit.

The three models that are used in the coaching

pipeline are mainly based on expert knowledge, with

some parts, such as some of the parameters, also esti-

mated from available data with statistical and machine

learning techniques. The elements of knowledge that

need to be encoded in a form that is available for rea-

soning are:

• The criteria that must be taken into account in

each particular model. What factors affect a spe-

ciﬁc phenomena and can be used to assess it?

• Collection of the possible outputs of each model.

For example, what are the appropriate coaching

actions in each given scenario?

• Deﬁnition of how the values of the input crite-

ria affect the assessment of the outputs of each

model. What rules govern the state of outcomes,

based on the states of the inputs?

For the purposes of formalisation of expert knowl-

edge and reasoning we use the multi-criteria mod-

elling methodology, which is presented next.

3 MODELLING METHODOLOGY

The three models from the coaching pipeline (as de-

scribed in Section 2.2) act as the main knowledge

representation and reasoning components of the sys-

tem. In ambient intelligence systems, the most com-

monly used knowledge representations are ontolo-

gies (Bouznad et al., 2017) as they offer rich knowl-

edge representation capabilities. Our initial design

of the modelling pipeline envisaged use of existing

problem domain ontologies (e.g., dogont

and uni-

versAAL

) for description of sensor readings and con-

textual information. A collection of simple rules that

would include elements from the ontologies was fore-

seen as the reasoning engine. However, the available

existing ontologies from this ﬁeld were found not to

be focused on concepts that are important in our appli-

cation, such as wearable sensors and coaching-related

concepts. Instead of expected minor adaptations, sig-

niﬁcant additional ontology development would be

needed. We therefore opted for a less demanding

modelling solution with a more elaborate reasoning

engine and weaker knowledge representation capabil-

ities: the hierarchical multi-criteria decision models

(MCDM), which are established in the ﬁeld of deci-

sion making and decision support.

Multi-criteria decision models (MCDM) are used

to evaluate, compare and study alternatives (Greco

et al., 2016). Typical examples of alternatives are,

for instance, cars, job candidates, ofﬁce locations, etc.

Usually the task is to select the most appropriate al-

ternative for speciﬁc goals, the one with the highest

http://iot-ontologies.github.io/dogont/

https://github.com/universAAL/ontology

Multi-criteria Modelling Approach for Ambient Assisted Coaching of Senior Adults

utility. Decision support with MCDM is based on a

hierarchical decomposition of the problem. Alterna-

tives are hierarchically decomposed into sub-concepts

(or aggregate attributes) and ﬁnally to a ﬁnite set of

basic attributes, which represent model inputs. Util-

ity of aggregate attributes is evaluated with functions,

which depend on the corresponding attributes located

on the lower levels of the hierarchy. In view of the

problem presented in this paper, the alternatives are

situation assessments, coaching actions and their ren-

derings in speciﬁc contexts. In case of coaching ac-

tions for example, their utility corresponds to the ﬁt-

ness of the coaching action in a given situation. Basic

attributes describe the context and the topmost aggre-

gate attribute represents the coaching action to be trig-

gered in such a context.

Due to our past experience, we used the method-

ology DEX (Bohanec et al., 2013), which is a quali-

tative multi-criteria decision modelling methodology

that is well established in practice and is supported by

freely available software, primarily by DEXi

. Sim-

ilarly as in many other methodologies of this kind,

its models have a hierarchical structure of variables,

which represent concepts relevant for solving the as-

sessment problem at hand. The lowest level concepts

are inputs for the model, which get hierarchically ag-

gregated into higher-level concepts up to the outputs,

which usually represent assessments of decision op-

tions. For example, in Figure 3 a hierarchical struc-

ture of such a model is shown in which the topmost

concept Mobility stand-up Coaching Action depends

on two sub-concepts: Ability to stand-up and Rela-

tive change. The latter further depends on Ability to

stand-up and Predicted ability to stand-up.

A distinctive characteristic of DEX is its focus on

qualitative modelling in which the inputs to the model

are qualitative values and the value functions (the

functions used for aggregation of criteria into higher

level ones) are rule-based, usually represented in tab-

ular form. In the exemplary model from Figure 3, the

value scales are shown to the right of each concept

and exemplary rules (a selection, not all) which are

used for calculation of the higher-level concept from

the lower-level concepts are shown in tables that over-

lay the arrows that indicate dependencies.

The qualitative nature of this methodology allows

the models to be transparent, which is particularly

useful in situations in which the operation of the

model must lend itself to human understanding. Au-

tomated AAL solutions are a prime example of such

a case. However, the use of qualitative values can

also represent a limitation in situations in which re-

lationships among the criteria are naturally numerical

http://kt.ijs.si/MarkoBohanec/dexi.html

Mobility_general

Ability to

stand-up

Ability to

walk

Ability to

climb/descend stairs

Ability to

stand

Number of

stand-ups

high

medium

low

Mode of

standing

with help

of hands

no help of

hands

Time

needed

fast

slow

... ... ... ... ... ... ...

... ... ...

very high

high

medium

low

very low

very high

high

medium

low

very low

Figure 2: Structure of the situation model for the Mobility

pipeline.

(summations, averages, etc.). Such relationships usu-

ally occur at the lower levels of models, thus, they

are commonly left out of the main model and are sep-

arately computed as inputs. If a direct inclusion of

such concepts in the models is necessary or beneﬁ-

cial, it can be done by using speciﬁc DEX methodol-

ogy extensions (Trdin and Bohanec, 2018;

Znidar

and Bohanec, 2010).

4 EXEMPLARY MODELS

In the following, we present selected models that were

developed for the Mobility domain pipeline. This

pipeline conducts an assessment of a person’s phys-

ical mobility — the ability to move. It is executed

daily and mostly uses the data of the past 24 hours.

The models were developed in collaboration of com-

puter modelling and domain experts, in this case do-

main experts for physical mobility and rehabilitation.

The structure of the situation assessment model

for Mobility is shown in Figure 2. The topmost con-

cept of Mobility general is decomposed into four dif-

ferent aspects of physical mobility. For one of them,

Ability to stand-up, further decomposition is shown,

while for the others, their sub-hierarchies are omitted

from the sketch.

Based on the input criteria, this model conducts

an assessment of a person’s mobility in general. As a

side-result we also get the values of all intermediate

concepts in the model, for example all the individual

abilities that constitute Mobility general. In the sub-

sequent model of the cascade, which selects a coach-

ing action, we usually consider the topmost concept

of the situation model as one of the inputs. However,

sometimes also intermediate concepts can be used as

criteria. In case of Mobility pipeline it makes sense to

separately coach for the general mobility, as well as

for the individual abilities to move, as coaching ac-

tions for each of these are in some cases different.

This way we can construct ﬁve coaching pipelines,

which all share a situation model, but have different

coaching action models.

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

Mobility_stand-up

Coaching Action

Relative

change

Predicted ability to

stand-up

Ability to

stand-up

large increase

increase

none

decrease

large decrease

send a message to a caregiver

send message to PU (positive only)

no action

Ability to stand-up Relative Change Mobility_stand-up

Coaching Action

very high large increase send message to PU+

very high increase send message to PU+

... ... ...

high decrease no action

high increase send message to PU+

high large increase send message to PU+

... ... ...

low decrease send message to

caregiver

... ... ...

very high

high

medium

low

very low

very high

high

medium

low

very low

Ability to stand-up Predicted ability to

stand-up

Relative Change

very high very high none

very high high increase

very high medium large increase

... … ...

Figure 3: Structure and a preview of decision rules of the coaching action model for the Mobility pipeline.

A coaching action model for the person’s ability

to stand-up is shown in Figure 3, which shows the

structure of relevant concepts and also some exem-

plary rules. There are only three coaching actions

possible: (I) send a message to a caregiver, (II) send

a message to the primary user (positive only) and (III)

no action. A suitable coaching action is chosen based

on the values of the Ability to stand-up and Relative

change. The Ability to stand-up is an input that orig-

inates from the situation model (see Figure 2), while

the Relative change is an aggregate concept that de-

pends on the values of the Ability to stand-up and the

Predicted ability to stand-up. The latter can be sim-

ply the most frequent value in a speciﬁc past time in-

terval or a result of prediction with machine-learned

tools, which can take into account past values as well

as abundant contextual information (weather, day of

the week, etc.).

The rules that are used to calculate the values

of Relative change and Mobility stand-up Coaching

Action are shown in the tables that overlay the de-

pendency arrows in Figure 3. The rules of Relative

change are simply expressing how different is the cur-

rent assessment of the ability to stand-up compared to

the usual values of a particular person. The idea of

the rules for Mobility stand-up Coaching Action is to

combine absolute and relative information in a way to

primarily consider the relative change, but differently

in case of some speciﬁc absolute values. For example,

a small relative decrease in mobility does not trigger

a coaching action if its absolute value is high.

5 DISCUSSION

In previous sections we presented an AAL problem

and a system that tackles it by employing qualitative

multi-criteria models in its knowledge modelling and

reasoning infrastructure. Use of such a design and

methodology is novel in the speciﬁc problem domain.

Its effectiveness will be objectively evaluated in the

scope of dedicated evaluation tasks in the correspond-

ing project, but here we outline some of its qualitative

characteristics and potential beneﬁts and drawbacks,

particularly in comparison to the entirely ontological

approach.

Besides very fast execution and feasibility of seri-

alization of the models (which makes them easy to

store and update), the main beneﬁt of MCDM ap-

proach, which was expressed as such also by several

stakeholders that were presented with the proposed

methodology, is the transparency and understandabil-

ity of the models and their operation. This is a very

important aspect in the targeted AAL domain. Some

of the possible reasons for their understandability are:

(I) use of (primarily) qualitative values and rules, (II)

Multi-criteria Modelling Approach for Ambient Assisted Coaching of Senior Adults

united representation of concepts and reasoning rules,

(III) limited representation of concepts.

The qualitative variables used in the models and

the corresponding (mostly) qualitative reasoning ele-

ments are transparent and easy to understand. There

is no black-box model or complex calculation func-

tion included from the level of model inputs onwards.

However, the qualitative nature of the models repre-

sents also a limitation when the natural representation

of the variables is numerical. There are methodologi-

cal extensions necessary (and some available) for this

kind of modelling, but the corresponding software

tools are only experimental. There is another negative

consequence of qualitative modelling: given a large

number of criteria to aggregate, the rule-sets can be-

come too large to grasp and manage. This is to some

extent alleviated by the use of hierarchical structures

and possibility of using rule summarization tools, but

could still represent a problem in some situations.

The MCDM models represent relevant knowledge

(concepts, their value scales and dependencies) and

at the same time function as reasoning tools (as they

incorporate operational rules). This importantly sim-

pliﬁes understaning of the system’s operation. On

the other hand, consequently there is no separation

of knowledge from its use, which limits the reusabil-

ity of the models for other purposes. Ontological ap-

proach is very different in this respect, with clearly

separate knowledge base and reasoning components.

In some sense, the MCDM models represent concepts

in a speciﬁc way (their speciﬁc aspect or interpreta-

tion), which is subjected to the model’s purpose or

output. This makes them more easily understandable,

but at the same time also less general.

The concepts in MCDM models are represented

with very limited information: name, description,

value scales, structure and rules that govern depen-

dencies of their values. Limited information and lim-

ited amount of relations among them (essentialy ony

one among each two) constrain the knowledge repre-

sentation and reasoning options to only the purposes

in line with the main purpose of the model. This is an-

other example of a generality versus understandability

tradeoff.

Difference among the ontological and MCDM ap-

proach to modelling is also in the assumption of the

open versus closed world. The ontologies assume an

open world and allow reasoning also about concepts

that are not represented in an ontology (e.g., having an

ontology of animals with only birds deﬁned, a con-

dition on not being a bird can be formed). On the

other hand, the MCDM models only contain what is

necessary for their reasoning purposes and for exam-

ple need options of none-existance explicitly deﬁned

(e.g., as a speciﬁc value in the value scale of a con-

cept).

The MCDM modelling methodology seems to

clearly have some important beneﬁts in tackling the

problem presented in this paper, but also some lim-

itations. Likewise, the ontological approach is very

general and powerfull, but has drawbacks for its use.

There is, however, also a possibility of combined

use and there were some attempts of it in some do-

mains (Bastinos and Krisper, 2013; Brahimi et al.,

2017). In AAL and similar problems, the two ap-

proaches could be used in combination on different

levels — ontological on the level closer to data and

MCDM on higher levels that are more important for

the people to ﬁnd them easy to comprehend. We in-

tend to experiment with such a hybrid approach, start-

ing with the use of ontologies for data fusion, in our

future work.

6 CONCLUSIONS

Knowledge modelling and reasoning parts of an AAL

system were presented in the paper, which differ from

the common purely ontological approach, but seem to

ﬁt well to the purpose of the application and the needs

of its users. We presented the architecture of the solu-

tion and the details of the applied multi-criteria deci-

sion modelling methodology, with a discussion of its

potential beneﬁts and limitations in the context of ap-

plication in the given problem domain. The character-

istics of this methodology somewhat limit its general

reusability, but make it very suitable for implemen-

tation of the reasoning components that are close to

human users.

ACKNOWLEDGEMENTS

The authors acknowledge the ﬁnancial support from

the Slovenian Research Agency for research core

funding for the programme Knowledge Technologies

(No. P2-0103) and the project SAAM – Supporting

Active Ageing through Multimodal coaching which

has received funding from the European Union’s

Horizon 2020 research and innovation programme,

Horizon 2020 SC1-PM-15-2017 Personalised coach-

ing for well-being and care of people as they age

(RIA), grant agreement no. 769661.

KEOD 2019 - 11th International Conference on Knowledge Engineering and Ontology Development

REFERENCES

Ayala, I., Amor, M., and Fuentes, L. (2013). Self-

conﬁguring agents for ambient assisted living ap-

plications. Personal and ubiquitous computing,

17(6):1159–1169.

Azimi, I., Rahmani, A. M., Liljeberg, P., and Tenhunen, H.

(2017). Internet of things for remote elderly monitor-

ing: a study from user-centered perspective. Journal

of Ambient Intelligence and Humanized Computing,

8(2):273–289.

Bastinos, A.

S. and Krisper, M. (2013). Multi-criteria de-

cision making in ontologies. Information Sciences,

222:593–610.

Bikakis, A. and Antoniou, G. (2010). Rule-based contextual

reasoning in ambient intelligence. In International

Workshop on Rules and Rule Markup Languages for

the Semantic Web, pages 74–88. Springer.

Bohanec, M.,

Znidar

c, M., Rajkovi

c, V., Bratko, I., and

Zupan, B. (2013). DEX methodology: three decades

of qualitative multi-attribute modeling. Informatica,

37(1).

Bouznad, S., Chibani, A., Amirat, Y., Lyazid, S., Prestes,

E., Sebbak, F., and Fiorini, S. (2017). Context-aware

monitoring agents for ambient assisted living appli-

cations. In European Conference on Ambient Intelli-

gence, pages 225–240. Springer.

Brahimi, L., Bellatreche, L., and Ouhammou, Y. (2017).

Coupling multi-criteria decision making and ontolo-

gies for recommending dbms. In 22nd International

Conference on Management of Data, COMAD, pages

08–10.

Chen, L., Nugent, C. D., and Wang, H. (2012). A

knowledge-driven approach to activity recognition in

smart homes. IEEE Transactions on Knowledge and

Data Engineering, 24(6):961–974.

Fides-Valero,

A., Freddi, M., Furfari, F., and Tazari, M.-

R. (2008). The persona framework for supporting

context-awareness in open distributed systems. In

European Conference on Ambient Intelligence, pages

91–108. Springer.

Greco, S., Figueira, J., and Ehrgott, M. (2016). Multiple

criteria decision analysis. Springer.

Herv

as, R., Bravo, J., Fontecha, J., and Villarreal, V. (2013).

Achieving adaptive augmented reality through onto-

logical context-awareness applied to aal scenarios.

Kara, M., Lamouchi, O., and Ramdane-Cherif, A. (2017).

A quality model for the evaluation aal systems. Proce-

dia Computer Science, 113:392 – 399. The 8th Inter-

national Conference on Emerging Ubiquitous Systems

and Pervasive Networks (EUSPN 2017) / The 7th In-

ternational Conference on Current and Future Trends

of Information and Communication Technologies in

Healthcare (ICTH-2017) / Afﬁliated Workshops.

Li, R., Lu, B., and McDonald-Maier, K. D. (2015). Cogni-

tive assisted living ambient system: a survey. Digital

Communications and Networks, 1(4):229–252.

Rafferty, J., Nugent, C. D., Liu, J., and Chen, L. (2017).

From activity recognition to intention recognition for

assisted living within smart homes. IEEE Transac-

tions on Human-Machine Systems, 47(3):368–379.

Trdin, N. and Bohanec, M. (2018). Extending the multi-

criteria decision making method DEX with numeric

attributes, value distributions and relational models.

Central European Journal of Operations Research,

26(1):1–41.

Yamada, N., Sakamoto, K., Kunito, G., Isoda, Y., Ya-

mazaki, K., and Tanaka, S. (2007). Applying ontol-

ogy and probabilistic model to human activity recog-

nition from surrounding things. IPSJ Digital Courier,

3:506–517.

Zavadskas, E. K., Naimavi

cien

e, J., Kaklauskas, A., Kruti-

nis, M., and Vaini

unas, P. (2008). Multi-criteria de-

cision support system of intelligent ambient assisted

living environment.

Znidar

c, M. and Bohanec, M. (2010). Handling uncer-

tainty in DEX methodology. In URPDM 2010: Pro-

ceedings of the 25th Mini-EURO Conference.

Multi-criteria Modelling Approach for Ambient Assisted Coaching of Senior Adults