USABILITY OF VISUAL DATA MINING TOOLS

Edwige Fangseu Badjio, François Poulet

ESIEA Recherche,

Parc Universitaire Laval-Changé

38, rue des Docteurs Calmette et Guérin

53000 Laval

Keywords: Usability, human-machine interaction, visual data mining

Abstract: Visual data mining is a field of research which needs knowledge from several domains: statistics, data

analysis, machine learning, artificial intelligence, human-machine interfaces, data or information

visualization. We are interested in visual data mining environment usability (man-machine interaction

quality). This paper investigates how usability aspects can be incorporated in visual data mining

environment so that usability can be taking into account during the design process of the tool without

prototype evaluation tests which are time consuming at design stage. We have defined and we present here a

set of criteria for improving visual data mining tools usability.

1 INTRODUCTION

In many research areas needing interfaces for

treatments needs, for a long time only technical

aspects were taken into account in the design of

systems. A lot of problems result from that process,

namely the end users difficulties to use the system in

order to realize their task. For this purpose, we

propose a new method for improving usability of

visual data mining tools. In the actual stage of our

research, this method consists of a set of criteria and

strategies for setting up each criterion. We have

made some experimentations of a subset of these

criteria but we just present definitions and strategies

here.

According to (Fayyad & al, 1996) data mining is

the non-trivial process of identifying valid, novel,

potentially useful, and ultimately understandable

patterns in data. Visual data mining consists of

visualization use as a communication channel for

data mining. For (Wong, 1999), visual data mining

lies in tighly coupling the visualizations and

analytical process into one data mining tool that

takes advantage of the strengths of all worlds. Visual

data mining is a recent research field.

In man-machine interaction research field, there

are usability evaluation techniques (Nielsen, 2000),

(Scapin et al., 1993). We took these techniques as a

starting point for the development of our method.

The techniques used to improve usability of software

in man-machine interface will be presented in the

third paragraph, after usability definition, followed

in fourth part of this paper by the method which we

worked out.

2 USABILITY DEFINITION

There are several usability definitions:

(IEEE, 1990) defines usability as how easily a

user can learn to operate, prepare inputs for, and

interpret outputs of a system or component.

For (ISO 9241-11, 1998) usability is the way

specified goals can be achieved with effectiveness,

efficiency, and satisfaction in particular

environments.

(Nielsen, 2003) define usability as a quality

attribute that assesses user interfaces ease of use.

The word "usability" also refers to methods

improving ease-of-use during the design process.

Nowadays, in most of research teams, there is a

usability evaluation test after the development of a

product. That evaluation makes the quality of

human-machine interaction evaluation possible.

254

Fangseu Badjio E. and Poulet F. (2004).

USABILITY OF VISUAL DATA MINING TOOLS.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 254-258

DOI: 10.5220/0002654202540258

 SciTePress

3 USABILITY EVALUATION

Many user interfaces guidelines can be used to

improve usability when applied in the design step.

These guidelines could not be applied in visual data

mining at all because visual data mining has some

particularities. We have to study the visual data

mining domain before the definition of criteria

which will be appropriate for improving usability.

We are going to present the visual data mining

process, before this presentation, in the next sub-

section, we present some methods for usability

design products.

3.1 Design for usability

This part presents some techniques used in order to

improve software usability. We have:

 Iterative design and evaluation: this

method involves design, evaluation and

redesign of the software,

 A method which implies the user in a

model or a designed product usability

evaluation,

 The last method implies the domain

expert in usability evaluation of a model

or a designed product.

These methods are time consuming. Our main

idea for our work is to take visual data mining tools

usability guidelines into account before starting the

design process. The study of visual data mining

process helped us for this purpose. Knowing that,

usability performance is measured relative to users'

performance on a given set of tasks, the measures

are success rates (whether users can perform the task

at all), the time a task requires, the error rate, users'

subjective satisfaction (Nielsen, 2001). We have

defined a list of usability metrics. Before explaining

these metrics we are going to present the visual data

mining process in the next sub-section.

3.2 The visual data mining process

There are different stages in the visual data mining

process:

1. selection of the data to be exploited,

2. choice of the visualization method to use

or passing to stage 4.,

3. data visualization,

4. choice of a visual data analysis method

among those proposed by the system,

5. visualization of the results,

6. evaluation of the results (which must be

easily understandable), followed by a

possible return at stage 1 or 2,

7. analysis of the results considered as new

knowledge,

8. return at stage 1 or stop.

Visualization is a stage of visual data mining

process, which provides graphical displays, and

animation on which investigator observations are

based. The user of such a system is not intended to

be a data mining or data analysis specialist but an

expert of the data domain. In order to perform a

significant analysis, the user must be helped

because, due to his lack of statistical background, he

may not be able to perform the right choices. The

usability criteria we define are intended to help

users.

4 USABILITY CRITERIA

4.1 Adaptability

4.1.1 Definition

The adaptability is the capacity of the system to be

adapted to the user's needs without any explicit

intervention from the user or its capacity to be

reacted according to the context and the needs and

preferences of the users.

4.1.2 Strategy

For setting up this criterion, we thought of the

possibility for the user to personalize his interface.

The purpose of user interface personalization is to

take into account the user strategies or preferences.

We also thought of the development of means

available for taking into account the experiment

level of the user (beginning, tested, occasional) like

his profile.

4.2 Curability

4.2.1 Definition

Curability is the user capacity to correct a non-

desired situation.

4.2.2 Strategy

Error rate, time required for a task execution are

factors usually taken into account during man-

machine tools usability evaluation. This criterion

recommends visual data mining tools designers to

display curatives means for the errors likely to occur

USABILITY OF VISUAL DATA MINING TOOLS

255

in their environment so that usability evaluation will

be successful.

4.3 Errors management

4.3.1 Definition

Error management refers to means allowing on one

hand to avoid or reduce errors, and on other hand to

correct them when they occur.

4.3.2 Strategy

It is a question here of setting up means to detect and

prevent errors. For example, all the possible actions

on the interface must be considered and more

particularly the accidental supports keyboard keys so

that not awaited entries are detected. Another case: if

the data analysis method chosen by the user is not

successful (method execution is not completed), it is

necessary to be able to propose another method to

the user without any system crash. The user must be

able to execute another algorithm for data analysis,

the method selection tool must be able to give not

only the most adequate algorithm to the problem

resolution but also the list of ranked algorithms.

Classification is done according to algorithms

evaluation criteria.

4.4 Feedback

4.4.1 Definition

Feedback recommends that after achievement of an

action, the system provides an answer to the user

informing him about the accomplished action and its

result, this, with a deadline for reply suitable and

homogeneous according to types of transactions.

4.4.2 Strategy

The visual data mining cycle can be time

consuming, depending on the size of the treated

data. Some information showing the user that the

treatments are going on, the progress report of the

treatments should be provided to the user.

4.5 Guidance

4.5.1 Definition

User Guidance refers to the available ways to advise,

orient, inform, instruct, and guide the users

throughout their interactions with a computer.

Good guidance facilitates learning and use of a

system by allowing the users: to know at any time

where they are in a sequence of interactions, or in

the accomplishment of a task; to know what the

possible actions are as well as their consequences;

and to obtain additional information (possibly on

demand). Ease of learning and ease of use that

follows good guidance lead to better performances

and fewer errors. (Bastien et al., 1993)

4.5.2 Strategy

In the visual data mining process, users have to

select an analysis method for the resolution of their

problem. Algorithm selection is an exploratory

process highly dependent on the analyst’s

knowledge of the algorithms and of the problem

domain. Our end users are not experts of data

mining or data analysis but an expert of the data

domain. When making choice of data analysis

method to execute, they have to execute the set of

available methods and select the most adequate

algorithm for the given problem. Running an

algorithm for a given task is time consuming,

especially when complex tasks are involved. Our

strategy here is to provide help to the user for the

selection of the most adequate algorithm for a given

task. A trivial solution for this problem is to

determine the best analysis algorithm. But, the No

free Lunch theorem (Wolpert et al., 1996) states that

if algorithm A outperforms algorithm B on some

cost function then there must exist exactly as many

other fonction where B outperforms A.

Given the wide variety of analysis method

available the selection of the right algorithm for a

problem is an important issue. There are some

research works from that field. For example we have

the StatLog and the METAL projects. As far as

METAL is concerned, several approaches have been

used. These approaches investigate the problem of

using past performance information to select an

algorithm for a given problem. For this purpose,

knowledge about past performance information are

stored and the authors use the approaches such as:

ontologies, case based reasoning, induction

algorithm to predict the performance of a given

algorithm on a task. For new cases these approaches

proceed by successive approximations and so lead to

a loss of information.

We chose a multi agents system for the

evolutionary needs of the system. We thus will be

able to use the assets of this paradigm, and more

particularly the autonomy of the agents as well as

the possibilities to distribute our treatments.

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

256

4.6 Multiplicity of returned

4.6.1 Definition

This criterion refers to the system capacity to

provide several visualization methods.

4.6.2 Strategy

Current state-of-the-art data visualization or

information visualization propose many data

representation techniques: geometric techniques,

icon-based, pixel oriented techniques, hierarchical

techniques, graph-oriented techniques, distortion

techniques, and dynamic or interaction techniques.

Everyone agrees on the fact that none of these

methods is better than the others in all cases. For the

same set of data, it is a question of envisaging

several possible methods of visualization.

4.7 On-line help

4.7.1 Definition

This criterion relates to the documentation

availability of the user.

4.7.2 Strategy

Data visualization is based on graphical methods. A

possible approach to set up this criterion is to make

appear contextual texts to the screen to inform the

user or to provide him explanations associated with

the visualisation method used or about the choice of

a split-criterion for decision tree.

4.8 Plasticity

4.8.1 Definition

This criterion refers to the system ability to

dynamically react to fluctuation on resources while

preserving ergonomic continuity.

4.8.2 Strategy

The transition from one stage to another one in the

visual data mining process must be perceived by the

system as well as the transition to the analysis of a

data base different from the previous one.

4.9 Training data re-use

4.9.1 Definition

Training data re-use criterion refers to the possibility

of pursuing a data mining process. Particularly, the

outputs of the system can be used like data input.

4.9.2 Strategy

Visual data mining process included preprocessing,

treatment, postprocessing. For example a decision

tree can be interactively constructed (instead of the

usual automatic approach). The first representation

of the data corresponds to the initialization of the

decision tree construction algorithm. The tree

growing can be stopped at any level of the

construction. An important and untreated aspect in

this type of environment is the possibility for the

user to go on with a previously stopped treatment.

Indeed, if the user stopped the visual data-mining

algorithm before the task is finished, the idea here is

to enable him to continue in the process without

having to start again at the initial stage.

5 CONCLUSION

As we have seen, data mining tools users can be data

mining experts, visualization experts or expert of the

data domain. Our visual data mining approach is

dedicated to the experts of the data domain. We have

presented some work about the usability of visual

data mining tools, in order to develop a software

program able to help that type of users, and to avoid

any redesign step generating waste of time and high

production costs (without however guaranteeing the

performances). To get this usability, we have

established criteria having to be taken into account

for a development of reliable and useful software.

These criteria are applied to visual data mining. We

have started the development of the corresponding

software program: a visual data-mining environment

dedicated to the data specialist. The criteria we have

defined can be used as a basis for other

achievements.

REFERENCES

Bastien, J. M. C., Scapin, D. L., & Leulier, C. 1999. The

Ergonomic Criteria and the ISO 9241-10 Dialogue

Principles: A pilot comparison in an evaluation task.

Bastien, J.M.C., Scapin, D.L. 1993. Critères

ergonomiques pour l’évaluation d’interfaces

USABILITY OF VISUAL DATA MINING TOOLS

257

utilisateurs. Rapport technique INRIA n° 156, Juin

1993, INRIA : Le Chesnay.

David H. Wolpert, William G. Macready, 1996. No Free

Lunch Theorems for Optimization, IEEE Transactions

on Evolutionary Computation, 1:67–82.

Hilario M. and Kalousis A. , 2002. Fusion of meta-

knowledge and meta-data for case-based model

selection. In Proceedings of the 5th European

Conference on Principles of Data Mining and

Knowledge Discovery (PKDD-01), Springer-Verlag,

Freiburg, Germany.

Nielsen J., 2000. Designing Web Usability: The Practice

of Simplicity, New Riders Publishing, Indianapolis.

Nielsen J., 2003. Usability 101,

(http://www.useit.com/alertbox/20030825.html )

Wong P. C., 1999 Visual data mining, IEEE Computer

Graphics and Applications 19(5), p 20-21.

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

258