Sunteți pe pagina 1din 8

International Conference on Advanced Information and Communication Technology for Education (ICAICTE 2013)

An Extension of the Eindhoven Classifica-


tion Model to the Educational Sector
Margarida Figueiredo1 Lídia Vicente2 Henrique Vicente1
Victor Alves3 JoséNeves3
1
Department of Chemistry & Évora Chemistry Centre, University of Évora, Évora, Portugal
2
Agrupamento de Escolas de Reguengos de Monsaraz, Reguengos de Monsaraz, Portugal
3
Department of Informatics, University of Minho, Braga, Portugal

Abstract rimental when unemployment is high and


labour markets are demanding more skills
This work presents an extension of the than ever before [1].
Eindhoven Classification Model to sort The reasons why young people have
adverse events root causes for the Educa- poor skills when they finish their school
tional Sector. Extended Logic Program- careers are many and diverse. Solving
ming was used for knowledge representa- this problem requires detailed knowledge
tion and reasoning with defective infor- of the causes that lead to this situation.
mation, allowing for the modelling of the The school dropout is one of those causes.
universe of discourse in terms of default It is a complex phenomenon, resulting in
data, information and knowledge. Indeed, economic and social losses, either to the
a systematization of the evolution process individual, family or the community to
of the body of knowledge in terms of which the person belongs. If the school
Quality of Information (QoI) embedded dropout is large in a country or in a de-
in the Root Cause Analysis was accom- veloped region, the consequences will be
plished, i.e., the knowledge representa- mainly damaging in terms of economic
tion and reasoning system proposed led to competitiveness and social environmental
a process of QoI quantification that al- degradation [2].
lowed the study of the event's root causes, In 2012 Portugal had a dropout rate of
on the fly. 20.8%, according to data existing in the
report of the official statistical office of
Keywords: Eindhoven Classification the EU. Spain, with 24.9%, and Malta,
Model; Knowledge Representation and with 22.6%, were the only European
Reasoning; Education; School Dropout. countries that in the past year had higher
values. Despite the decrease (from 38.8%
1. Introduction to 20.8% between 2005 and 2011), this
rate is still far from the national goal of a
Education is a powerful driver of devel- 10% dropout from secondary schools, in-
opment and one of the soundest instru- tended for 2020 [3].
ments for reducing countries poverty. Al- The school is a complex and multifac-
though there has been great progress in eted system where varied types of ad-
the last decade, a large number of young verse events occur. An adverse event may
people that finish their education did it be defined as the failure of a planned ac-
without acquiring basic skills necessary tion to be completed as intended or the
for work and life. This is particularly det- use of a wrong plan to achieve an aim,

© 2013. The authors - Published by Atlantis Press 596


and includes problems in practice, rela- verse events [5, 6]. This approach assumes
tionships, procedures and systems. The that humans are fallible and that errors
most effective way to prevent adverse are to be expected in every organization,
events is to attack directly their causes. so it is necessary to concentrate efforts on
Preventing the adverse events' root causes the conditions under which individuals
improves significantly the educational work and try to build defences to avert
success. Thus, the proposed model will errors or to mitigate their effects. Assign-
focus primarily on preventing the adverse ing codes to the causes of each adverse
events' root causes. The model planned event, it is a practice that is useful for
serves as the formal foundation to an ad- tracking and trending.
verse event reporting and learning com- The first step when using the ECM
putational system. based classification system is to identify
the root causes that result in a specific
2. The Computational Model adverse event. These root causes are sub-
sequently classified according to the clas-
An educational version of the Eindhoven sification model. To do so, a causal tree is
Classification Model (ECM), with the ex- built and techniques of Root Cause
tensions and adaptations for the educa- Analysis (RCA) are applied. Once the root
tional field and its causal tree, used to causes are identified, they may be used to
classify the adverse events’ root causes in provide a more realistic view of how the
school dropout, is presented. The theo- system really works, as well as to con-
retical foundation is based on an exten- tribute to the creation of effective and
sion to Logic Programming, in terms of a lasting solutions.
revision of its knowledge representation
and reasoning mechanisms. Undeniably, 2.2. The Extended Eindhoven Classi-
the introduction of explicit negation in fication Model
this universe allowed for the development
of a process of quantification of the Qual- The Extended Eindhoven Classification
ity of Information (QoI) of the predicates Model (EECM) was adapted from the
extensions that make a logic program or ECM, presented in the previous section.
theory, making possible to study the To apply this model to the educational
event's root causes and to generate alerts sector, extensions were developed for
and recommendations on quality improve- each category of the original model.
ments in the didactic process. These extensions allow fitting each cate-
gory into the educational arena and pro-
2.1. The Eindhoven Classification vide a broader view of the events that
Model (ECM) may occur and the degree of complexity
of this field. Thus, the classification proc-
The ECM was originally developed to ess becomes easier and more efficient.
manage human error in the chemical proc- Figure 1 presents the EECM flow chart
ess industry [4], and was then applied to and the codes assigned to classify each
various other industries, such as steel indus- adverse event. For instance, in the origi-
try, energy production and in healthcare. nal model, the adverse events classified
The Eindhoven Classification Model – as “Human behaviour – Knowledge-based
Medical Version consists of 20 (twenty) errors” (HKK) occur due to “the inability
codes, divided into four categories fre- of an individual to apply existing knowledge
quently used in a medical environment to to a novel situation”. In the EECM, this
classify the underlying causes of the ad- definition was extended by saying that
the events classified under this category are

597
Start

Technical Yes No No No No
External? Design? Construction? Materials?
Factor?
No Yes Yes Yes

TEX TD TC TM

Organizational Yes No Knowledge No Management No No


External? Culture?
Factor? Transfer? Priorities?
No Yes Yes Yes Yes

OEX OK OM OC

Surrounding Yes No Labour No


Requests?
Factor? Market?
No Yes Yes

SR SL

Human Yes No Yes No


External? K-B? Knowledge?
Behaviour?
No Yes No Yes

HEX HKK

Yes No No
R-B? Qualifications? Co-ordinator?
No Yes Yes

HRQ HRC

Yes No No
S-B? Slips? Tripping?
Yes Yes

HSS HST

Multi-Failure Yes
MFC
Criteria?
No

Unclassifiable = X

Fig. 1: The Eindhoven Classification Model for Education.

due to “difficulties in execution, interpre- measure of confidence in any qualifica-


tation or reporting procedures”. Some of tion process susceptible to be handled by
the adverse events falling into this cate- the system. Since an event may only oc-
gory are “poorly executed procedures, in- cur due to the combination of more than
complete procedures and procedures one cause, and a different event may
poorly validated”. come about due to two or more causes,
The causal trees taken on by the origi- taken separately, in the original model
nal ECM, set that the recognition of the AND-gates and OR-gates are used to em-
event’s root causes and its mental picture, body these two possibilities in the causal
is done under a hierarchical structure. On tree. The usual situations may also in-
the other hand, once one has to deal with clude the case where only one cause leads
incomplete and even contradictory infor- to the occurrence of a certain event. In
mation, an Extension of Logic Program- any case the adverse events’ origins are
ming (ELP) was used to knowledge rep- known, i.e., there are sure about the
resentation and reasoning, in order to get events’ grounds. Beyond these situations,
a truth value in the interval 0…1 as a it may happen that the causes of an event,

598
action or decision are unknown; may be ability to reason with incomplete infor-
known that certain views are the source mation, since in the real world complete
of a given event, but may not be sure information is hard to obtain, even in the
what are the event grounds; or it is not most controlled situations. The idea be-
allowed to know the origin of a given hind default information is the ability to
event (e.g. due to internal policies of the make assumptions or to jump to a plausi-
schools). Therefore, it is proposed the use ble conclusion, derived from a knowledge
of “unknown” and “forbidden” operators, base in the absence of information to the
to allow the representation of unknown contrary. The derived information is de-
values of an infinite set of values, un- feasible, because in light of new informa-
known values of a given set of values, tion the conclusion may need to be re-
and values not allowed or forbidden. The tracted, i.e., we are in the presence of
information contained in each causal tree non-monotonic reasoning [8]. A suitable
is then represented in ELP by the exten- logic is needed, one that permits the rep-
sions of a predicates set, being also used resentation of incomplete, inconsistent
as a formalism to quantify the causal and default information and supports non-
treeś QoI (see Section 2.4). The QoI al- monotonic reasoning. In a classical logi-
lows the identification of the causes that cal theory or logic program, the proof of a
should be taken into account, in first theorem (here understood as a question
place, and how this hampers all the clas- submitted to the classification system) the
sification process. The information ob- outcome is a truth value, namely true or
tained in this way to the RCA enables false [9]. ELP introduces another kind of
automatic report generation with im- negation, strong negation, represented by
provements in the recommendations. Fig- the classical negation sign ¬. In most
ure 2 presents the application of the situations, it is useful to represent ¬A as a
EECM to the adverse event School Drop- literal, if it is possible to prove ¬A. In
out. In the source of this event there is a ELP, the expressions A and not A, being
great diversity of reasons. It is possible A a literal, are extended literals, while A
that only one situation might be enough or ¬A are simple literals. Intuitively, not
for the event to occur or, perhaps, it may A is true whenever there is no reason to
be necessary a combination of several believe A, whereas ¬A requires a proof of
factors. The causal trees should include the negated literal.
all possible causes and aim to be a ge- Every program is associated with a set
neric representation of the problem. For a of abducibles, which may be seen as hy-
particular occurrence of the event, its potheses that provide possible solutions
causes will fall on a branch of the tree. or explanations of given queries, being
given here in the form of exceptions to
2.3. Knowledge Representation and the extensions of the predicates that make
Reasoning the program. The issue is providing ex-
pressive power for representing explicitly
In the past few decades, many non-classical negative information, as well as to directly
techniques for modelling the universe of describe the closed world assumption for
discourse and the reasoning procedures of some predicates, also known as predicate
intelligent systems have been proposed, in circumscription [10]. Three types of an-
addition to classical ones [7]. Of particular swers to a given question are then possi-
interest to this work are the techniques to ble, i.e., true, false and unknown. The rep-
deal with incomplete, inconsistent, con- resentation of null values will be scoped
tradictory, default and forbidden informa- by the ELP. It is possible to consider three
tion [8]. Intelligent systems require the

599
School Dropout

or

Personal Community
Unknown Forbidden

or or

Community
School
Student Family Outside the
Community
School

and or Unkn or
own

Lack of Lack of
Failing in Inappropriate Lack of Ethnic or High Number
Report not
Economic Report not Extra-School Qualifications Alternatives to
School Behaviours Parental Cultural of Students
written
Motives reviewed Requests Requirement the Standard
in Classes Interest Reasons per Class
Curriculum

Extra-school
Requests

HKS HRB HEE HEC HEC SR SL OM OEX

Fig. 2: Extended Causal Tree for the Adverse Event School Dropout.

types of null values: the former will allow  The professional that recorded the
for the representation of unknown values, school dropout situation only re-
not necessarily taken from a given set of ported family reasons. It is not pos-
values, the middle one will represent un- sible to be constructive, concerning
known values taken from a given set of the action or truth-value to consider.
possible values, and the latest will define However, it is false that the action
values that are not allowed or forbidden. or decision could be different. This
Taking the example of the adverse event situation suggests that the lack of
School Dropout (Figure 2) it might repre- knowledge may be associated to a
sent all the possible situations according set of possible known values – un-
to the following setting: known value in a finite set of values
(in this case there are three possi-
 It is known that the student is in bilities, i.e., economic motives, lack
school dropout because he/she is of parental interest or eth-
failing in school – known value; nic/cultural reasons);
 It is only known that the student is  And finally, namely due to internal
in school dropout. In this case who policies of the educational institu-
reported the adverse event did not tion, it is not permitted to know the
know which actions or decisions causes of a given event – forbidden
led to the event occurrence – un- or not allowed values.
known value;

600
Using ELP, as the logic programming
language, it is now possible to set a pro- 1
cedure given in terms of the extension of
a predicate called demo: question, answer
→ [0,1]. Given a question (Q), it returns a latest
case former
solution based on a set of assumptions,
case
where question indicates a theorem to be
proved and answer denotes a truth value middle
(see Program 1; true (1), false (0), being 1 term case
unknown (U) in the range of the truth
values ]0,1[). 1

demo(Q,T) ← Q Fig. 3: The embedded QoI with respect to the


demo(Q,F) ← ¬Q question Which are the actions or decisions
demo(Q,U) ← not Q ∧ not ¬Q that led to the adverse event occurrence?.

Program 1: Extension of meta-predicate demo. It is now possible to measure the QoI


associated to the question put in context,
2.4. Quality of Information in terms of a logic program that endorses
procedures of action_or_decision, which
The Quality of Information (QoI) with may be given in the form Which are the
respect to the extension of a generic pred- actions or decisions that led to the adverse
icate p may be analysed in different con- event occurrence?. The shaded n-slices
texts and measured in the interval [0,1]. (here n is equal to 3 (three)) of the circle
When the information is known; when the denote the QoI.
information is unknown; when the infor-
mation is unknown but can be taken from
3. Discussion
a set of values. If the information is
known the QoI for the extension of predi- Based on the formal approach referred to
cate p is 1. For situations where the value above, an adverse event reporting and
is unknown the QoI is given by: learning system was introduced. Indeed,
lim N   0 N  0
1 to the professionals of education and
N mostly for all the education institutions,
Finally, if the information is unknown but this approach may bring some advantages.
can be derived from a set of values, the After the adverse events being registered,
QoI is set in terms of 1/Card, where Card similar to what happens in other reporting
denotes the cardinality of the abducibles systems, the analysis process becomes
set for p. easier, more expedite and reliable. Un-
As an example the QoI associated with doubtedly, with the recourse to ELP,
the information about the RCA of the ad- leading to an on the fly measurement of
verse event School Dropout, for the first the QoI of the logic terms used in the
three cases present in the previous section, process of judgement (in terms of a theo-
is depicted in Figure 3 and is given in the rem to be proved), the human interven-
form: tion in the analyze process is only neces-
sary to approve the recommendations,
Vaction_or_decision (former case) = 1; causes and events that may need attention.
Vaction_or_decision (middle term case) = 0; It also caters for the credibility and the
Vaction_or_decision (latest case) = 0.33.

601
measurement of the efficacy of the im- plished. A way to solve the representation
plemented strategies and actions. problem of defective information was
Although the causal classification of presented, adequate for evaluating the
events is sometimes time-consuming and QoI in such situations. It was also pre-
difficult, with the development of a ge- sented a computationally feasible formal
neric causal tree for each possible event, tool to measure the value of QoI. With
the increase in time consuming is on the this approach to RCA and classification it
initial phase of the model enforcement. was possible to identify the causes, ac-
The QoI allows the ordering of causes, tions and decisions that may lead to the
identifying the ones that should be taken adverse events and define the strategies to
into account in the first place. In the ge- prevent them.
neric tree it is necessary to consider all
possible causes, rather than the most 5. Future Work
probable or usual ones. The information
obtained is useful in identifying possible In the future an Adverse Event reporting
trends and areas requiring further investi- and learning System applied to the Edu-
gation. cational Sector (AESES) will be devel-
The conceptualized logic model offers oped. The AESES will comprise 3 (three)
the means for knowledge extraction, pro- core modules, making it not only a sys-
viding the identification of the most sig- tem for adverse event registration, but
nificant causes and suggestions of changes also a learning system. The Adverse
in the organization policies and procedures Event Reporting Forms in the Educa-
of schools, subject to formal proof. In- tional Sector (AERFES) module will pro-
deed, the creation of an inference system vide a Web interface for adverse event
in support of the logical model enables registration. The effort on this interface
the generation of reports with strategies will be focused in its usability. The event
for quality improvement on the fly, where registration will be made by professionals
a quality measure of the system is on of the educational institutions (e.g. teach-
oneś confidence on the results, in terms ers, administrative staff, auxiliary staff),
of the QoI. by the students and by their parents,
through predefined forms adapted to each
4. Conclusions user profile. The Adverse Events Man-
ager Reports in the Educational Sector
The main contribution of this work is to (AEMRES) module will be also Web
be understood in terms of the evaluation based and aims to enable the analysis of
of the QoI in the RCA and the possibility the adverse events recorded by AERFES,
to address the issue of incomplete infor- based on the Extension of the Eindhoven
mation, through the use of an Extension Classification Model (EECM). The sys-
to Logic Programming (ELP) in the con- tem will provide an individual report for
struction of causal trees. ELP was used each adverse event recorded, which will
for knowledge representation and reason- include all its details and the extended
ing with defective information, catering causal tree obtained using the EECM. The
for the modelling of the universe of dis- AEMRES module will also provide charts
course in terms of incomplete, inconsis- with statistical information about the im-
tent, forbidden and default data, informa- pact, place of occurrence, type of form
tion and knowledge. A systematisation of and type of event recorded. Finally, the
the body of knowledge’s evolution about Adverse Events Knowledge Manager in
QoI embedded in the RCA was accom- Educational Sector (AEKMES) module

602
will use the data from the system data- [3] http://europa.eu/rapid/press-release_IP-
base to create a Knowledge Base (KB), 13-324_en.htm.
which although had been given in terms [4] T.W. van der Schaaf, “Near miss re-
of ELP, will be rewritten to productions porting in the chemical process indus-
in the logic programming language try: An overview”, Microelectronics
PROLOG [11], based on the EECM. Reliability, vol. 35, no. 9-10, pp.
From the KB other reports relevant to the 1233-1243, 1995.
improvement of the scholar environment [5] T.W. van der Schaaf, “PRISMA-
may be generated, always with the assur- Medical: a brief description”, Eind-
ance of data reliability and credibility, by hoven University of Technology,
taking into account its QoI. Faculty of Technology Management,
Patient Safety Systems, 2005.
[6] S. Rodrigues, P. Brandão, L. Nelas, J.
Acknowledgements Neves, V. Alves, “A logic program-
ming approach to medical errors in
This work is funded by ERDF - European imaging,” International Journal of
Regional Development Fund through the Medical Informatics, vol. 80, no. 9,
COMPETE Programme (operational pro- pp. 669-679, 2011.
gramme for competitiveness) and by Na- [7] F. Sheridan, “A Survey of techniques
tional Funds through the FCT - Fundação for inference under uncertainty” Arti-
para a Ciência e a Tecnologia (Portu- ficial Intelligent Review, vol. 5, no. 1,
guese Foundation for Science and Tech- pp. 89-119, 1991.
nology) within project FCOMP-01-0124- [8] M.L. Ginsberg, Readings in Non-
FEDER-028980. monotonic Reasoning, Morgan
Kauffman Publishers Inc, USA, 1987.
[9] U. Hustadt, “Do we need the closed-
References world assumption in knowledge rep-
resentation?”, In: Baader, Buchheit,
[1] http://siteresources.worldbank.org/EDU Nutt (Eds.), KI’94 Workshop, Saar-
CATION/Resources/ESSU/Education_ brüken, Germany, 1994.
Strategy_4_12_2011.pdf. [10] J. Neves, “A logic interpreter to
[2] T. Andrei, D. Teodorescu, B. Oancea, handle time and negation in logic
“Characteristics and causes of school data bases”, Proc. of the ACM’84
dropout in the countries of the Euro- on the fifth generation challenge, pp.
pean Union”, Procedia – Social and 50-54, 1984.
Behavioral Sciences, vol. 28, pp. 328- [11] I. Bratko, Prolog Programming for
332, 2011. Artificial Intelligence, Pearson Edu-
cation Canada, Canada, 2011.

603