# INTRODUCTION he research in this paper is an investigation on how to apply data mining rules [1,2,4], especially clustering algorithms, to e-learning usage data to dynamically create just in time learner models. Author ? ? ? : NRI Institute of Technology and Manage ment, Baraghata, Jhansi Road, Gwalior-474001, INDIA. E-mails : drkishansharma2006@rediffmail.com, priyainfo005@gmail.com , sharmanitm.06@gmail.com Make Sense of Usage Data: E-learning systems are used for computerbased education and they have widespread use in many domains. The usage data collected from the forum for each learner includes: Messages posted in the forum, question messages posted in the forum, answering messages posted in the forum, messages accessed by the learner, messages mostly navigated by the learner. # Issues of Using Usage Data: With rich usage data collected from e-learning systems, we try to make sense of this data by applying data mining techniques. There are some challenging issues that need to be navigated: among patterns found from data mining techniques, patterns are useful in an e-learning system, determine that a pattern is useful or not, predict learners' behaviors based on the usage data. # OBJECTIVE In this paper, we did some proof of concept research to study the above issues. We studied the relationships between usage data and learner [6] characteristics and behaviors. This resulted a six layers model to create learner models. This is a dynamic model created by applying clustering techniques on the usage data collected from the real system. We implemented a test system to collect data and to create results. Two experiments have been used to evaluate and compare the results of the test system. # III. PROBLEM DEFINITION Some patterns found from the usage data, also called metrics and measurements to represent learners' characteristics, seem to be clearly useful in building learner models. Other patterns show promise to describe learners' behaviors, but remain unproven. Deferent clustering algorithms produce various results. Selection and determination of data mining algorithms and associated parameters will play an important role in creating learner models. Pre-computation is necessary if anything like just in time modeling is to be achieved, and has been implemented in our test system Global Journal of Computer Science and Technology Volume XI Issue XVII Version I 31 IV. # LITERATURE REVIEW In order to more easily discuss the [1,2] current state of web based e-learning systems, educational data mining and my own research, it is useful to _rest look at the history that has brought educational research and data mining technologies together. It focus on web based educational theories such as adaptive intelligent learning and learner models, data mining algorithms, and educational data mining research. # b) Learning Content Management Systems Developing a course [3]to be taught on the Internet is difficult because it requires the system to do a combination of things: publishing content on web pages, supporting tools for self learning, and providing assessments of learning performance Some good commercial LCMS systems include Blackboard (Web CT ), Virtual-U and Top Class, etc. Open source LCMS include iHelp, a Tutor and Model, etc. c) iHelp iHelp is an e-learning system developed[5,7] by the Advanced Research in Intelligent Education Systems (ARIES). iHelp is made up of a number of web based applications designed to support both learners and instructors throughout the learning process1. The main components of iHelp are asynchronous iHelp Discussion forums, synchronous iHelp Chat rooms, the iHelp Learning Content Management Systems (also called iHelp Courses), iHelp Share and iHelp Lecture. _ iHelp Chat: This chat room provides workspaces for learners to have synchronous communication with one another and with their instructors and teaching assistants. _iHelp Courses: This LCMS system provides tools to support full on-line courses and is designed for distance learning. It provides learners with a portal to multimedia course content. _iHelp Share: This is a collaborative learning tool to share information relevant to courses among learners. _iHelp Lectures: This system provides multimedia lectures to learners so that learners can write messages and comments, make notes and tags on video clips, so that all learners can share this information. Like other LCMS systems, iHelp collects and stores all information, such as personal information, pedagogical results, learners' interaction data, etc. into a database. These data are the source data for our project, as we will discuss in the. V. # RESEARCH CONTRIBUTION AND FUTURE DIRECTION The goal of this research has been to show that just in time learner models can be created from analyzing learners' online tracking data. This approach consists of clustering raw data, selecting pedagogical applications and applying data mining methods. This has led to measurements and metrics that can be calculated for each individual learner to represent that learner's characteristics and behaviors. # a) General Comments on the Two Experiments From the two experiments' results, some measurements seem to be useful in building just in time learner models; [1] some measurements only show promise to be leading in the right directions; some measurements have not found much support. The expert experiment, in which experts observed and evaluated learners as the third party, shows much more positive results compared to the self evaluation experiment, in which learners evaluated themselves. (i) A goal of this research has been to compute learner [3] models just in time as we need them instead of keeping static learner models. We do not use the historical learner models in our computations, instead recomposing the metrics and measurement based on the current available fact data and raw data. In this way, any changes in learners will be automatically recognized in the form of new measurements. Through two experiments, we have shown that just in time computations are possible and in some cases they lead to useful measurements. # c) Top Down Top down computation of [2,6] measurements promises to allow the calculation of results quite rapidly when compared to computations using a pure bottom up data mining approach. In our approach, the first step is to decide the purpose of the applications such that we can figure out the necessary metrics and measurements to support the Applications. The second step is to find the raw data and retrieve the fact data to support the measurements. The last step is to mine the fact data to find patterns that can be used in creating formula that can later be used to directly calculate the measurements based on the available fact data. # VI. FUTURE WORK AND DIRECTION The main drawback to the two experiments is that the number of evaluators is too low to get statistically significant results. It is easier to have more learners involved in the self evaluation than to get more experts in the expert evaluation. However, we need to improve and design a better self evaluation questionnaire to attract more learners to participate while at the same time get better questions. More involvement of instructors in the design of the questionnaire may be helpful. We have selected four metrics and 15 measurements in the two experiments. Then we need to show how we can use these measurements to build just in time learner models in actual pedagogical applications. Because of the relatively high correlation coefficient results in the two experiments, we can at various time points, apply classification algorithms to predict learner behaviors. A promising direction may be to keep old predictions and combine this information with the latest updated predictions to improve the accuracy of the just in time learner model. # VII. # Correlation 32![Figure 3.2 : Six layers model](image-2.png "Figure 3 . 2 :") ![a) E-Learning E-learning is naturally associated with computer based learning, especially to be used in distance learning. The term e-learning is also called by some researchers e-training, online instruction, web-based learning, web-based training, web-based instruction, etc. The main advantages of e-learning are edibility and convenience. Learners can work at any place and at any time with an Internet connection for most e-learning environments.](image-3.png "") Table 5.1,5.2 : Summary of expert experimentb) Just In Time Model32experiment clued three measurements from the activitylevel metric: navigating context, read in discussionforum, chat activity; two measurements from the socialtendency metric: presence and social tendency; andone measurement from the knowledge tendency metric:usage instructors to observe and evaluate learners'learning behaviors as in traditional class rooms. Theresults support that the instructors [5] at least will have ahelpful tool to dynamically observe and evaluatelearners'performanceinonlineeducationenvironment. Six measurements had negative results inthe expert experiment [4,6], with lower accuracy valuesor lower correlation coefficient values. Thosemeasurements in the learning style metric especiallyhave lower values in both accuracy and coefficientvalues.© 2011 Global Journals Inc. (US) AccuracyCorrelationAccuracyCoefficient<0.400.4-0.6>0.6Coefficient<0.400.4-0.6>0.6<0.4210<0.43110.4-0.61110.4-0.61200.6-0.80110.6-0.80100.6-0.81460.6-0.80002011October332nd International Conference on Educational DataMining, pages 1{10}.5. © 2011 Global Journals Inc. (US) E-Learning by Time Dynamic Model Using Data Mining © 2011 Global Journals Inc. (US) E-Learning by Time Dynamic Model Using Data Mining ## ACKNOWLEDGEMENTS The authors are very thankful to the referees for their useful suggestions for the presentation of the paper. * Detecting symptoms of low performance using production rules JBAgapito SSosnovsky AOrtigosa 2 nd International Conference on Educational Data Mining 2009 * Acquiring background knowledge for intelligent tutoring systems CAntunes 1st International Conference on Educational Data Mining 2008 * Inferring unobservable learning Variables from students' help seeking behavior IArroyo TMurray BWoolf CBeal Lecture Notes in Computer Science 3220 2004 * A comparison of student skill knowledge Estimates EAyers RNugent NDean 2009