Abstract — E-learning systems with an endemic impact in academics greatly rely on learner categorization for prolific delivery of learning contents. However, current techniques may not fully comprehend the implicit as well as explicit characteristics of learners having very subjective consideration of academic aspects without dynamically taking into account the complete learners’ perspectives in rightly categorizing the learners. In this paper, a machine learning based dynamic and adaptive technique named LCHAIT (Learner Categorization based on Hybrid Artificial Intelligence Techniques) is proposed for learner categorization with an objective focus on learner attributes. A supervised mode of learning was employed on a labelled data set, modelled through a “Learner Ontology”, having diverse learners’ profiles with implicit and explicit attributes pertinent to learners’ perspectives of demographics, academics, inclinations and behaviours. A comparative analysis of LCHAIT with three other machine learning techniques (Fuzzy Logic, Case Based Reasoning, and Artificial Neural Networks) is also presented. The efficacy of all techniques was empirically measured while categorizing the learners based on their profiles through metrics of accuracy, precision, recall, f-measure performance and associated costs. These empirical quantifications assert LCHAIT as a better option than contemporary techniques as exhibited by greater accuracy of performance metrics. The novelty of the proposed technique is signified through preprocessing of feature attributes, quality of data, training of machine learning model on more relevant data, and adaptivity. Index Terms— Learner Categorization, Semantic e-Learning, Artificial Neural Networks, Fuzzy Logic, Case Based Reasoning I. INTRODUCTION E-learning systems [1, 2, 3] are playing a pivotal role in transforming information societies into knowledge societies through widespread delivery of didactic contents by educating human communities. These systems are driven by three key stakeholders: i.e. Instructor, Learner and System Administrator. Instructors, educationists with a supervisory role, design the learning contents, exercises/assignments and exams to educate and assess learners. Learners on the other hand, are consumers of learning contents, undergo the learning cycle of learning, assessments and corrections to master certain course(s). System administrators, with a role of facilitators, harness the platform for instructors and learners in performing their respective roles. Here, the focus of our work “This work is submitted on 26 December 2016”. Sohail Sarwar is with the Department of Computing Iqra University Islamabad, Pakistan (e-mail: Sohail.sarwar@seecs.edu.pk). Raul Garcia Castro is with Universidad Politecnica de Madrid, Spain, (e- mail: rgarcia@fi.upm.es). Zia Ul Qayyum is working with University of Gujrat, Pakistan (e-mail: zia.qayyum@uog.edu.pk). Muhammad Safyan is working with Government College University of Lahore Pakistan (e-mail: m.safyan@seecs.edu.pk) Farrukh Latif is working in Department of Computing, Bahria University, Islamabad (e-mail: farrukhlatif01@gmail.com ) is pertinent to the learner and specifically to the problem of Learner Categorization. Steepness in the “learning curve” of learners defines the degree of effectiveness exhibited by an E-learning system throughout the learning cycle (learn, assess, adapt) [1]. So it may be asserted that the success of an E-learning system is greatly dependent upon the learner which in turn depends upon the delivery of learning contents to the learner. Any irrelevance in presenting contents to the learner would not only hamper the learning process but would also result in a waste of time and resources incurred in delivering these contents. Therefore, a typical “one size fits all” approach [1] may not fully comprehend the learner’s capacity to learn while presenting learning contents. Hence different factors need to be considered while offering the learning contents to the learner instead of presenting the same contents to every learner. These factors are learner’s academic performance, learning style, aptitude, background knowledge, and term-wise performance during the course. Once the picture of the learner’s ability is clear by rightly categorizing the learner, learning contents may be offered keeping in view the learner’s ability. This categorization has potential to offer twofold benefits i.e. what contents to be offered to the learner for first time and adaptivity of contents (performance-based sequencing/re-sequencing of contents during the course). Different techniques have been proposed for learner categorization [1, 2, 3, 5, 9] that categorize the learners exploiting the “unsupervised mode” of learning for classification. These techniques may not comprehend both implicit (such as score in exams, CGPA and performance in Pre-Requisite courses, age, and region) and explicit characteristics (such as Pre-Test score, learning style, aptitude and personal details) of learners in categorization; specifically, academic aspects are considered very subjectively [2, 3, 6, 9] in contrast to objective ones. Moreover, techniques considering these subjective academic aspects ignore learners’ behavioral and demographic perspectives. Some techniques [15, 16] sort the learners into good and bad learners only, which is not amenable to a true depiction of learner’s categories. Few learner categorization techniques, after categorizing the learners, do not take advantage of reusing information of newly categorized learners for future classifications. This prevents machine learning techniques from being dynamic and adaptive to cater new scenarios dynamically. Lastly, few techniques [9, 10] claim to target the semantic web but formal and explicit descriptions of learners using ontologies seem missing. Keeping the facts above in view, an adaptive and dynamic learner categorization technique named LCHAIT (Learner A Hybrid Intelligent System for Effective Learner Categorization in Semantic e-learning Systems Sohail Sarwar, Raul García-Castro, Zia Ul Qayyum, Muhammad Safyan, Farrukh Latif International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 12, December 2016 516 https://sites.google.com/site/ijcsis/ ISSN 1947-5500