| Web-based educational technologies allow educators to study how students learn (descriptive studies) and which learning strategies are most effective (causal/predictive studies). Since web-based educational systems are capable of collecting vast amounts of student profile data, data mining and knowledge discovery techniques can be applied to find interesting relationships between attributes of students, assessments, and the solution strategies adopted by students. The focus of this dissertation is three-fold: (1) to introduce an approach for predicting student performance; (2) to use clustering ensembles to build an optimal framework for clustering web-based assessment resources; and (3) to propose a framework for the discovery of interesting association rules within a web-based educational system. Taken together and used within the online educational setting, the value of these tasks lies in improving student performance and the effective design of the online courses.; First, this research presents an approach to classifying student characteristics in order to predict performance on assessments based on features extracted from logged data in a web-based educational system. We show that a significant improvement in classification performance is achieved by using a combination of multiple classifiers. Furthermore, by "learning" an appropriate weighting of the features via a genetic algorithm (GA), we have successfully improved the accuracy of the combined classifier performance by another 10--12%. Such classification is the first step towards a "recommendation system" that will provide valuable, individualized feedback to students.; Second, this project extends previous theoretical work regarding clustering ensembles with the goal of creating an optimal framework for categorizing web-based educational resources. We propose both non-adaptive and adaptive resampling schemes for the integration of multiple clusterings (independent and dependent). Experimental results show improved stability and accuracy for clustering structures obtained via bootstrapping, subsampling, and adaptive techniques. These improvements offer insights into specific associations within the data sets.; Finally, this study turns toward developing a technique for discovering interesting associations between student attributes, problem attributes, and solution strategies. We propose an algorithm for the discovery of "interesting" association rules within a web-based educational system. The main focus is on mining interesting contrast rules, which are sets of conjunctive rules describing interesting characteristics of different segments within a population. In the context of web-based educational systems, contrast rules help to identify attributes characterizing patterns of performance disparity between various groups of students. We propose a general formulation of contrast rules as well as a framework for finding such patterns. Examining these contrasts can improve the online educational systems for both teachers and students---allowing for more accurate assessment and more effective evaluation of the learning process. |
