| Summary: | 碩士 === 元智大學 === 資訊工程學系 === 101 === Data mining methods include fuzzy theory, decision trees, regression analysis, data clustering, inference rules, neural and other teaching activities. Data mining methods have been shown to find out information among large amounts of data for further interpretation and analysis by experts, to assist human in problem-solving and decision-making. This research applied two data mining methods: correlation analysis and decision tree to predict whether a student correctly answers a question or not.
This study used the data of Taiwan Assessment of Student Achievement to establish correlation model and decision tree model to predict student answer performance. The study computed precision, recall and F value of two methods to evaluate the prediction accuracy of correlation model and decision tree model. The evaluation results revealed: 1) Decision tree analysis technique can be applied to predict each question while correlation analysis technique can be applied to some questions because some questions did not have generated association rule. 2) Generating associative rules requires severe conditions,, so the prediction accuracy rate of correct answer was over 90 % and the relative the recall was down to 20 % 3) There is no obvious correlation between the predication accuracy of the association analysis and the discrimination of the question while there is statistically positive correction between the predication accuracy of decision tree analysis and the discrimination of the question. 4. The average F value of predicting correct answer condition of association analysis was 0.54 while the F value of decision tree was 0.82; The average of F value of predicting the wrong answer was 0.57 while the F value of decision tree was 0.6; the overall average of F value of decision tree 0.78 was better than that of association rules (0.56), 5. Decision tree can be generated for each questions while association analysis may not be applied to some questions.
|
|---|
