Automatic classification of learning objects through dimensionality reduction and feature subset selections in an e-learning system

This Research paper focus on the design and development of intelligent, personalized mobile agent for learning object classification and retrieval. We have used JADE (Java Agent Development Environment) platform to launch, migrate, classify and retrieve the learning content based on the customized query by a peer learners in an virtual e-learning environment like MOODLE. In turn the agent collects the user query and migrates into different Learning Object Repository, (LOR), learn and interact with them and retrieves the Learning Objects(LO) along with its Learning Object Metadata (LOM). From the retrieved learning objects the learner applies the machine learning algorithm for the purpose of classifying the LO based on his usefulness/interestingness/learning objective and etc., The retrieval agent has been trained by the machine learning algorithms and then this trained agents now migrates into different such LORs applies the classification in that respective place and retrieves only the relevant and ranked learning objects to the learner based on the learners pre classification techniques and interest. The main drawback in the case of traditional Client-Server based learning environment, the learner client portal takes the query of the learner, and retrieves all the result set irrespective of usefulness/interestingness of the learner. Here the learner spent time to get the irrelevant or non interested content along with the required set itself. The bandwidth also used to the transfer such uninterested content is waste. These two drawbacks can be avoided or rectified through the mobile agent which posses the capability of classification based on feature subset selection techniques like Infogain, entropy and machine learning algorithm then migrates into the different LOR for collecting the relevant LO. As soon as it is once migrated into the LOR, it applies the machine learning algorithm first and then classifies the result to eliminate the non interested content not to be retrieved and transported to the learner side. The result of experiment using there different classifier, TF-IDF, Bayesian and Fuzzy classifier for the design and development of trained document classifier are Presented. The feature subset selection algorithm can be integrated with this to improve further accuracy of classifications.(future work).