In designing intelligent web based educational systems different student needs and preferences should be taken into consideration. Personalization of a system usually results in an increase in its effectiveness, which can be measured by the degree to which learning outcomes are achieved. However, taking into account the individual requirements of each learner and adjusting the system to their needs may be very costly in cases of large numbers of students. Finding groups of learners with similar preferences seems to be a good solution which allows for differentiating the system in compliance with the needs of group members. Learner preferences depend on characteristic traits such as cognitive features, including dominant learning styles. Students modeled by learning style dimensions can be assigned into groups by using unsupervised classification. The performance of educational software depends on the quality of obtained clusters, which is connected with the degree of similarity between students who belong to the same groups. The proper choice of clustering algorithm is crucial for receiving the expected results. Evaluation of the applied clustering technique can be performed through validation of the obtained student clusters taking into account the resemblances between their members.
The aim of the chapter is to compare different validation techniques of cluster analysis for grouping students characterized by their dominant learning styles. Different validation criteria applied for disparate clustering algorithms are considered. On the basis of experiments which use real student data as well as artificially generated datasets, certain indications concerning the most appropriate validating criteria and the best clustering schema are presented.
The chapter is organized as follows. A research review concerning intelligent e-learning systems, student grouping as well as cluster analysis and validation techniques is presented in the next section. Following this, the clustering techniques and validity criteria that have been applied for student grouping are described in more detail. In the following section student models as well as problems connected with the application of unsupervised classification for finding student groups are presented. Then, on the basis of experiment results, some indications are given concerning the best validity criteria as well as the best clustering methods. Finally, future research directions and a conclusion are presented.