Research On Classification Problem Based On Improved Fuzzy Case-based Reasoning Algorithm

As a method to solve given problem based on solved cases, case-based reasoning represents significant prospects for improving the effectiveness and accuracy of unstructured decision-making problems. Similar problems have similar solution that is the core assumption of Case-based reasoning system.This paper proposes a fuzzy case-based reasoning method to represent the utility of cases to a new problem due to the completeness and globalism of fuzzy rules on extracting domain information compared to the traditional distance similarity modeling. Due to the simplicity and effectiveness of WANG algorithm on generating fuzzy rules, this paper modifies the WANG algorithm to obtain fuzzy rules from data directly to improve the performance of system on classification problem. The feature attributes are considered as the inputs of fuzzy rules to determine whether and which extent a solved case is useful to the given problem. By applying the method on the standard UCI data sets to verify its effectiveness on classification problem. The simulation result presents that the method proposed in this paper not only has the advantages such as short training time, easy parameter choosing and easy algorithm programming achieving, but also has good performance when there are a limited number of cases available.Considering a large number of data with more than ten or hundreds of features which leads to "curse of dimensionality" with respect to the fuzzy rule number as well as the heavy computational costs, we apply the genetic algorithm as a combinatorial optimization problem to identify relevant feature attributes out of a big feature set to reduce the complexity of our system. The simulation on WINE data sets shows that the Fuzzy case-based reasoning combined with feature selection not only reduce the complexity of classification system greatly, but also improve the classification accuracy. By applying to the leakage fault diagnosis of hydraulic chamber, we verify the feasibility of our method on diagnosing and classifying the fault in the real world. Compared to the diagnosis performance of BP network on the same data sets, the method proposed in our paper shows better result on identifying the fault classification.