Homogeneous Grouping of E-learners Using Ensemble Clustering

Despite the individual differences of learners such as their abilities, goals, knowledge, learning styles and backgrounds, most of the electronic learning systems has presented an equal learning content for all of the learners. This is happening while producing a specialized content for the individuals. Increasing appliances of artificial memory in teaching the adaptation learning systems will require recommended teaching methods which are appropriate to the learner’s individual differences. In order to grouping learners based on their learning styles in their own similar groups, we are presenting a new method in this text. This method is mainly about combining the result of clustering methods which is certainly reducing choosing an unreliable method. Meanwhile it is preventing method`s complication which is because of using simpler and more useful clustering algorithms that subsequently will cause a better result and it may happen due to the fact that different methods will overlap each other’s defections. In this article we are using Felder- Silverman learning style which consist of 5 dimensions: processing (active-reflective) , input (visual-verbal) , understanding (sequential-global) , perception (sensing-intuitive) and organization (inductive-deductive). Firstly, proper behavioral indicators to different learning style dimension of Silverman-Feedler will recognize and then based on these behaviors learners will be able to be groups by one of these 5 methods. In the case of evaluating the proposed method, utilizing the c++ programming electronic teaching period information is necessary. Learner members of experiment environment were 98 ones which were extracting the expressed indicators connected to their network behaviors in 4 dimensions of Perception , process , input and understanding of Felder- Silverman model. On the other hand students were asked to fill the questionnaire forms and their learning styles were calculated between 0-11 and then based on the behavioral information they were being grouped. We are using 5 clustering grouping methods k-means , FCM , KNN , K-Medoids and SVM to produce ensemble clustering in generation step and co-occurrence samples or majority votes were used in Integration step. Evaluating the results will require the followings : Davies-bouldin index , Variance index , and gathering purity index. Due to the fact that the expressed methods are not able to indicate automatically the best cluster, clustering 3,4,5,6,7 clusters were using this method. And with calculating Davies-bouldin index the best cluster in each method were selected. In FCM each data were contributed to the cluster which has the most dependence to that . Numerical results of Davies-bouldin index have shown that ensemble clusters have the exact accumulation clusters among the others. Clustering variance in different size is indicating that ensemble clustering has the most accumulation and the least dispersion and also purity-gathering results has shown that proposed grouping method has the ability to gather learners with the similar style in each cluster and has a better efficiency compared to the others. So with this idea while maintaining simplicity, more accurate results based on the Davies-bouldin index , Variance index , and gathering purity index is obtained. Due to the importance of high accuracy and high speed and low computational complexity in the clustering methods, instead of a more complex approach, combining the weaker and easier clustering methods, better and more accurate results reached.