Neural Network-based Data Mining Platform For The Design And Implementation

In the 21st century, with the development of Information-Society and the popularization of Internet, the information quantity has been increasing so quickly that it has become more and more urgent to find some effective tools aiming at analyzing and disposing of the data that we need. With the development of Neural Network in the 1980s and the popularization of KDD (Knowledge Discovery in Databases), the technique of Data mining emerges as the times require. Data mining is a process of selecting, exploring, analyzing and modeling in the enormous amount of data, with the purpose of finding unknown rules and associations among the data for people to find some valuable results. Neural Network aims at building an auto adaptive dynamic system through finding a mathematics model to simulate the activity of human's brain nerve, and has the feature of strong adaptive ability and faster speed of parallel operation. In the field of Data mining, some kinds of Neural Network are used to the classification procedures, such as Perceptron, BP (Back Propagation) and RBF (Radial Basis Function Neural Network); some are used to the unsupervised clustering procedures, such as SOM (Self-Organizing feature Map) and RBF; some are used to the statistics prediction, such as RBF. Neural Network is very compatible for descriptive and predictable mining process. GA (Genetic Algorithm) is such an algorithm that it is a stochastic search algorithm and simulates the process of organism's natural selection and inheritance. GA has several excellent features: hidden parallelism, global search, and easy to combine with other methods.Java is a powerful oriented object and cross-platform programming language, and it is especially compatible for the web's programming. Nowadays, a lot of companies and organizations have developed their software based on Java. As the standard of Java's enterprise development, J2EE (Java 2 Platform Enterprise Edition) provides the developers with a complete and powerful middle platform. MVC (Model-View-Controller), which was proposed by Trygve Reenskaug, now has become a popular design pattern used in J2EE. In the paper, I organize the architecture from two aspects: theory and software development. First, it introduces the concept of Data mining and the application of Neural Network in the former; then introduces four kinds of Neural Network in detail. That's including: Perceptron, BP and RBF, which are used to the classification procedures; and SOM, which is used to the unsupervised clustering procedures. At the same time, it gives each Neural Network an application example in order to show their functions more intuitively. Then, the paper introduces the knowledge of GA, and focuses on the issue of how to use GA to optimize the structure and the primary weight of BP by giving my analysis and opinion; at last, I make a test to show that it is more effective to train BP's Network combined with GA.In the section of software development, it first introduces the online Data mining project-MinerOnWeb; then according to the standard flow of software engineering, it introduces how to develop the Data mining platform based on Neural Network which is used the J2EE architecture in detail. The platform is composed of four function models: Getting data source, Training Neural Network, Saving the mining model and Applying the model. The process of development is connected to the four kind of Neural Network we have indicated closely. Meanwhile, the paper gives some pictures to show the effectiveness of using each Neural Network to make the process of Data mining on the platform.