Knowledge Discovery from E-Learning Activities has been useful for detecting patterns in e-learning data. Despite the possible problems of converting continous numeric data to discrete value data, improvement of interpretation capabilities has been demonstrated. Modelling learning dimensions as a combination of Web event activities enhanced the detection of the student learning styles. The knowledge discovery from e-learning Web data found useful knowledge (of global or particular content) on academic performance of the students at the Universidad Politécnica Abierta (UPA). Among the findings are the following: (i) Events of synchronous interactivity, such as chats, forum participation, and events of asynchronous interactivity empower the student academic performance; (ii) In the courses with grades, academic student performance could be improved by motivating students to have course achievement. Some students show good values for the different event activities, including exercise practice, but do not have evaluations. General results of the research were well evaluated by academic experts taking into account the validity, novelty, and simplicity of the knowledge. All these knowledge of global and particular contents could be used to improve the e-learning system in different aspects. Strategies to encourage interactivity between students, strategies to design an assessment methodology that reinforce the student learning styles detected, and global improvements of different components of the e-learning system towards a more distributed interactive learning could be proposed. Considering the findings of knowledge, a preliminary set of strategies was outlined. Future work As a prototype, the study has yielded encouraging results on the application of knowledge discovery to e-learning analysis. Nevertheless, in order to obtain a complete application of this analysis is necessary to complement the data warehouse with more variables. Thus, the complexity of the analysis of the research topics can be more realistically modelled. Among these variables could be gender, age, location, enrolment date, likes, and dislikes, and so on. Besides of teacher’s data, such as the course’s survey results and research topics. Those variables would be collected through questionnaires, or transferring automatically from databases. The part of teaching styles of the Felder’s learning framework or another educational model has to be incorporated in the proposed methodology. The tuning of the learning and teaching styles to obtain a good performance in the outcome of the process would be modelled. Depending on the mixture of learning and teaching styles, several adaptations of the pedagogical e-learning resources could be made. The results of the enhanced model could be used to adapt teaching methodologies, including the critical aspect of the assessment style, or in general to improve the e-learning system, balancing distributed passive learning (DPL) and distributed interactive learning (DIL). The semantic information and the implicit informatics ontologies defined by clusters descriptions, decision trees, and learning styles conceptions found in the research, could be used to implement a knowledge-based system and/or a standard ontology of the studied domain. The ontology could be used to exchange and reuse the knowledge and it would make easy to increase, foster, and update the knowledge obtained from the Web e-learning activities. The use of a Web ontology language and standard data interchange formats would make possible the approach to the semantic Web. Future reseArch dIrectIons The chapter has discussed the knowledge discovery in e-learning considering several subjects as: e-learning Web activities, data preprocessing, data mining techniques, knowledge evaluation, learn-
Knowledge Discovery from E-Learning Activities ing and teaching styles, pedagogical innovation, andinformatics ontologies. The balance between interactive and personal activities is a critical factor for e-learning systems (distributed passive learning (DPL) and distributed interactive learning (DIL)). An interesting area of research is the proposal of new e-learning activities or determining the suitable mixture of those activities considering, for instance, contents, multimedia resources, and ubiquitous networks. The quality of the discovered knowledge is directly proportional to the cleanness and relevance of data. E-learning processes could generate a lot of useless information, so efficient algorithms to filtering and summarizing data; to resolve inconsistencies, to estimate missing data, and solving data heterogeneity are valuable for the knowledge discovery approach. Pattern recognition is a wide area that includes many kinds of machine learning algorithms. Independent component analysis (ICA) algorithms, as applied in the present chapter, have yielded important results in areas as image filtering and segmentation, brain to computer interface, and electrocardiographic diagnosis. Recently the mixture of ICAs has emerged as a flexible generating model to arbitrary data densities using mixtures of Gaussians or Laplacians distributions or nonparametric distributions for the components. Those ICA mixtures could be used to model data or knowledge in the Web. Usually the evaluation of the knowledge is made by experts. Nowadays, aspects as novelty or interestingness are estimated by novelty detection algorithms. Those algorithms could be used by intelligent agents in the Web in order to make decisions considering user behaviours. In the field of e-learning it is a novel approach. Recently, second level patterns in data mining have been studied. Those approaches have been used in proteinand ADN research, where the results of a first level of data mining conforms a huge knowledge domain. In Web applications that kind of methods would be useful. In addition the automatic conversion of knowledge or semantic information obtained by Web mining techniques, represented by structures as the decision trees, to ontologies would make possible the exchange and reuse of the domain knowledge. Methodologies to create hierarchical structures of patterns are suitable to create ontologies; to make that possible, conversion procedures to translate statistical information to standard data interchange formats are needed. All of those approaches may contribute to develop the semantic Web. AcknowLedgment Special thanks go to the Universidad Politécnica Abierta personnel for giving the Web data andinformation about the virtual campus. This work has been supported by Spanish Administration under grant TEC 2005-01820. reFerences Barker, K., Trafalis, T., & Rhoads, T.R. (2004). Learning from student model. In System and Information Engineering Design Symposium (pp. 79-86). Beck, J.E., Jia, P., Sison, J., & Mostow, J. (2003). Predicting student help-request behavior in an intelligent tutor for reading. In 9th International Conference on User Modelling (pp. 303-312). Beck, J.E., & Woolf, B.P. (1998). Using a learning agent with a student model. Lecture Notes in Computer Science, 1452, 6-15. Bezdek, J.C., & Pal, S.K (1992). Fuzzy models for pattern recognition: Methods that search for structures in data. New York: IEEE Press. Borst, W.N. (1997). Construction of engineering ontologies for knowledge sharing and reuse. University of Twenty, NL-Centre for Telemática and Information Technology.