IDENTIFICATION OF A BUGRULE BY BAYESIAN NETWORK MODELING
 
Ohmori, T. and Shigemasu, K., The University of Tokyo, Japan
 
The purpose of this study is to design and implement a Bayesian diagnosis system to identify the cause of the mistakes committed by elementary school students. The system is based on the probabilistic network model, which incorporates the psychological process model of the problem solving. The current system is targeted for fraction calculation and factorization. This system automatically diagnoses the optimal test items and administers them to the subject. Each time the subject gives the answer to the computer system, correct or wrong, the system calculates the posterior conditional probability of each possible cause of the mistakes based on the accumulated data. New aspects of this network system is the use of the MCMC method to efficiently calculate the posterior probability. Real data were obtained by applying the system to the elementary students, and they showed that the system was effective the cause of the mistakes of the students.