Recently there has been significant advances in the use of wavelet network methods in various data mining processes, With the extensively application of many databases and sharp development of Internet, The capacity of utilizing information technology to fabricate and collect data has improved greatly. It is an imperative problem to mine useful information or knowledge from large databases or data warehouses. Therefore, data mining technology is urbanized rapidly to meet the need. But data mining often faces so much data which is raucous, disorder and nonlinear. Providentially, ANN is suitable to solve the before-mentioned problems of DM because ANN has such merits as good vigor, flexibility, parallel-disposal, distributing-memory and high tolerating error. This paper gives a detailed discussion about the relevance of ANN method used in DM based on the analysis of all kinds of data mining technology, and especially lays stress on the categorization Data Mining based on RBF neural networks. Pattern classification is an important part of the RBF neural network function. Under on-line environment, the training dataset is variable, so the batch learning algorithm which will generate plenty of surplus retraining has a lower efficiency. an suitable metric for imbalanced data is applied as a filtering technique in the context of Nearest Neighbor rule, to improve the classification accuracy in RBF and MLP neural networks This paper deduces an incremental learning algorithm from the gradient descend algorithm to improve the blockage. ILA can adaptively adjust parameters of RBF networks driven by minimizing the error cost, without any surplus retraining. Using the method projected in this paper, an on-line cataloging system was constructed to resolve the IRIS classification problem.