Study On Uncertain Reasoning Theory And Knowledge Discovery

The research on uncertain information processing is an important work in current artificial intelligence research field. In order to process uncertain information, many mathematical tools and methods have been developed, including fuzzy sets theory, bayesian belief network, D-S evidence theory and rough set theory. These theories and methods deal with the different aspects of uncertain information. They have been the key theories and methods in current data mining and knowledge discovery. This paper focused on the uncertain information processing, developed these theories and methods further, studied the integrated system based on different methods dealing with uncertain information, and achieved the following research results.1. Presented one fuzzy inference model with fuzzy probability factor. Prove that this kind of fuzzy inference model is an universal approximator, and proposed one method identifying this fuzzy model from training data set. In this fuzzy model, the certainty factor of each rule is interpreted as a fuzzy conditional probability of rule consequent given rule antecedent. The simulation in time series prediction problem shows that the prediction performance of the fuzzy inference model with fuzzy probability factor is always better than the classic Mamdani-type fuzzy inference model without fuzzy probability factor.2. One simple Naive Bayes network is extended to fuzzy Naive Bayes network. Based on fuzzy Naive Bayes, fuzzy Naive Bayes classifier is studied, and this classifier is applied to some famous machine learning examples. Here, fuzzy Naive Bayes network is an important tool identifying other intelligent systems, since from this fuzzy Naive Bayes, not only the classic fuzzy if-then rule can be extracted, but also the fuzzy D-S belief structure can be extracted.3. Studied the fuzzy inference model with fuzzy probability factor based on fuzzy Naive Bayes and genetic algorithm. This fuzzy model is applied to a complicated control simulation梑acking truck and time series prediction problem successfully. This fuzzy model uses fuzzy Naive Bayes to extract the classic fuzzy if-then rules and the certainty factors of rules, at the same time, it uses genetic algorithm to optimize the fuzzy partition of input space and output space.4. Studied the identification problem on one kind of fuzzy classification system with weights, this fuzzy classification system with weights is applied to some famous machine learning examples and achieves the satisfactory results. This fuzzy classifica-tion system with weights not only uses fuzzy Naive Bayes to extract the classic fuzzy if-then rules, but also uses fuzzy Naive Bayes to extract the weights in the antecedent of rule.5. Studied one kind of general fuzzy system based on fuzzy D-S belief structure, proposed one inference method using the firing degree in the antecedent of rule to affect the membership function shape of focus elements in the fuzzy D-S belief structure in the consequent of rule. When the membership function of each focus element has the same power, this fuzzy model become the well-known Sugeno-type fuzzy inference model. The belief of each focus element in fuzzy D-S belief structure is determined by fuzzy Naive Bayes network.6. Proposed another inference method for the general fuzzy system based on fuzzy D-S belief structure. In this inference method, the firing degree of antecedent of rule affects the belief of each focus element in the fuzzy D-S belief structure in the consequent of rule, and one more new focus element "unknown" is induced. The analysis shows that the combinatorial explosion is alleviated largely. Here, The belief of each focus element in fuzzy D-S belief structure is determined by fuzzy Naive Bayes network.7. Taking into account the importance of past experiences and knowledge, this paper proposed the evidence combining theory based on a prior knowledge. This theory has rigorous probabilistic foundation and some algebra properties. And the evidence combining theory without a prior knowledge...