Bayesian Networks Is Applied In The Diagnosis Of Cognitive

Traditional Measurement theory only describes the examinees with certain scores or certain level of proficiency, not providing sufficient referent information for remediation. The phenomenon that examinees with the same score or equal ability have different cognitive structure can not be explained in the context of traditional test theory. Nowadays, people are not satisfied with tests that can only give a summative assessment. In contrast, they require test that can provide formative information for assessment. As a new tendency of measure theory, cognitive diagnostic assessment (CDA) emphasizes the role that cognitive psychology played in test construction. CDA is designed to measure specific knowledge structures and processing skills in students so as to provide information about their cognitive strengths and weaknesses. Therefore, CDA attracts more and more attention from the researchers and test developers. The main task of CDA is to diagnose examinees’ cognitive states through their performances on the test, and then classify all the examinees to specific cognitive attribute modes. Around this task, various diagnostic models and methods have been developed. Tatsuoka’s rule space model (RSM) and its variation-attribute hierarchy model (AHM) developed by Leighton have great influence in this circle. Q matrix that emphasized in both models above plays a vital role in diagnostic test construction and classification. As along as the cognitive attribute hierarchy is known, Q matrix can be attained smoothly. In this sense, a reasonable cognitive attribute hierarchy is quite important to CDA.Bayesian network is based on Bayesian theory and graph theory. It can express the independent and dependent relationships between variables in qualitative and quantitative way, through its acyclic graph and probability distributions. On the one hand, it is suitable for computer processing; on the other hand, it is straightforward and convenient to set a model. The establishment of a Bayesian network is a learning process with data. Bayesian networks are used to show the structure of the problem in much more research fields. Bays classifier is a special kind of Bayesian network and its classification efficiency is confirmed in many areas.This research made a try to use the theory and techniques of Bayesian networks into the CDA. This research can be divided into two parts. The first part is the Bayesian structure study using the data of258students, in order to get the attribute hierarchy relationships. The overcome can be used to validate and connect the former hierarchy proposed by the subject experts. The second part is the diagnostic classification using naive Bays classifier and Tree Augmented naive Bays classifier. It included both simulation research and empirical research. Both of them show that naive Bays classifier and Tree Augmented naive Bays classifier have a good performance in diagnostic classification and have a wider application aspect in CDA.