Data mining in intelligent tutoring systems using rough sets

Abstract

Data mining aims at searching for meaninghl injormation like patterns and rules in large volumes of data. Our objective is to mine the data of Intelligent Tutoring Systems (rrS). n e s e are tutoring systems which offer the ability to respond to individualized student ne&. An qweriment was conducted over a lesson for binary relatbns. Students’ answers to questions at the end of the lesson were collected. Data mining was implemented to extract important nrles @om the data (students’ answers) and hence the student can be directed to which parts of the lesson he should take again, thus heking to adopt the brtoring systems to each student individual needs nree approaches are applied to detect the decision nrles based on the Rough Sets and the Md$ed Rough &s. n e s e approaches provide a poweijid foundation to discover important structures in data. These approaches are unique in the sense that they onb use the injormation given by the data and do not rely on other model assumptions. f i e results obtained were in the form of rules that showed what concepts the student understood and which he did not understand depending on which questions he answered correct and which questions he answered wrong. Also some questions of the quizzes were found to be useless. It was concluded that data mining was able to extract some important patterns and rules >om the students’ answers which were hidden before and which are helpfir1 to both the students and the expert. Data mining is a set of methods used as a step in the Knowledge Discovery (KD) process to distinguish previously unknown relationships, rules and patterns within large volumes of data [l]. One of data mining tasks is &scription i.e. to describe databases in terms of patterns which human can understand and make use of In our research, we are trying to mine databases resulting from Intelligent Tutoring Systems (ITS). These are Computerbased tutoring systems which achieve their intelligence by representing pedagogical decisions about how to teach as well as information about the learner. This allows for greater versatility by altering t k system’s interactions with students. Intelligent tutoring systems have been shown to be highly effective at increasing student’s motivation and performance [2]. 0-7803-8575-6/04/$20.00 02004 IEEE The goal of data mining in ITS is to automatically assess student knowledge of the concepts underlying a tutorial topic, and use this assessment to direct remediation of knowledge. It does not require any knowledge about the subject being taught [3]. Thus our main objective is to investigate the application of data mining to provide a reliable way to determine a student knowledge status i.e. what a student does and does not know during the course of instruction. Once student knowledge can be assessed automatically without human intervention, computer4ased educational system can be individually tailored to each student’s leaming needs. This research investigates the application of rough sets as a method of data mining and knowledge discovery in ITS. Rough Sets approaches provide a powerful foundation to reveal and discover important structures in data They have been shown to be very effective for revealing relationships within imprecise data, discovering independencies among objects, removing redundancies and extmcting decision rules [4]. Three approaches were applied in this research two different approaches of traditional Rough Sets [5] and one approach of Modified Rough Sets [6]. The rules extracted were evaluated to test their quality using complexity measure [7]. The data sets on which the three approaches were applied came from a trial run of a binary relations lesson in North Carolina State University (NCSU) with a group of students taking Discrete Mathematics during the past three years. The binary relations lesson is taken over the NovaNET network which is a computer-based educational network that evolved from the first computer-based educational network, PLATO. NovaNET is a system-quality network with response times in fractions of seconds. Its programming language, TUTOR, is particularly effective for writing educational lessons, including modes for answer judging and integrated help. Lessons on NovaNET integrate text and graphics, and can also be linked to Internet resources, including audic-visual presentations The second section of this paper describes the rou& sets algorithms implemented. The third section describes the three different methods for the extraction of decision rules. The fourth section is for data analysis and the fifth section deals with the results obtained. Finally, the last section presents t he conclusions and future work. 131.