Model pentaksiran pembelajaran pelajar menggunakan kaedah hibrid

Abstract

In the advent and merging of information and communication technology (ICT) in education, more and more computer aided learning systems are being developed in Malaysia to assist teaching and learning. However, most of these applications are still lack of individualized attention as compared to learning from the human instructor. In order for a learning system to provide individualized attention, the system must have the capability to assess and reason about the student. This has prompted to a research in the field of student modeling that aims to assess and reason the student's mastery and efficiency level in a drill and practice type of learning Programming Technique 1 i.e C programming language. Assessment and reasoning the student performance is not an easy task, especially when it involves many attributes and various skills. In this research, there are four factors required to measure two student's output. The four factors are the average marks, time spent, attempts, and help needed. The two output to be measured are the mastery and the learning efficiency level. The knowledge of the human expertise is always acquired to determine the criteria of the students' learning behavior and the decisions about their level of mastery and efficiency. Unfortunately, most of their information is always vague and incomplete. In order to overcome these problems, this research studies hybrid artificial intelligent techniques that are the neural-fuzzy and the roughfuzzy techniques. The back propagation neural-fuzzy approach is used to solve the problem of vagueness in the decision made by the human expertise. By training the neural network with selected patterns that are certain &om the fuzzy inference, the network is then can recognize other decisions that are previously not certain. To overcome the problem of rules incompleteness, a complete fuzzy rule base is formed by combining all possible term sets of the fuzzy input and associated it with the decisions made about the student's performance and learning efficiency obtained from the back propagation neural-fuzzy method. It is found that the total number of fuzzy rules being constructed depend directly on the definition of the size of the input fuzzy term set. The more the size of the input fuzzy term set, then the more the fuzzy rules are formed. Too many fuzzy rules will occupy more computing time and space. Therefore, the rough-fuzzy approach, which involves the fuzzy logic and the reduction process of the rough set theory, is introduced to reduce the size of the fuzzy rules in the knowledge base optimally. Several experiments have been conducted on the proposed model's prototype, and it has successfully assessed the mastery and efficiency level of the student and classified the student into categories as determined by the human instructor