Multi-Dim Data Analysis Based On Sequence Constructive Neural Network

Sequence Constructive Neural Network-SCNN, is a new model of neural network, which is built based on neurons dynamic constructive technology. This method is more suitable for the critical requirement of machine learning, which is the dynamic, active procedure but not the static, passitive one. In this paper, principles, constructive methods and implications are studies, and also made some researches on the multi-dimensional data analysis based on the SCNN.As we all known, new model and method is a critical problems which is should be solved in the intelligence machine study. Artificial Neural Network has been widely applied in the fields, such as pattern reorganization, data analysis. Currently, with information technologies have been developed with high speed, a great mount data has been generated. New development of technology gives new problem for neural network. Classical methods which are based on full connection or its improved method are difficulties for multi-dimensional data analysis requirement which is fixed with special structures.Classical Neural Network has problems such as difficult definition in hidden layer, slow conversation rate, difficult training and multi-dimensional data analysis, and no idea affection in real application. Therefore, SCNN method is discussed combining the status of information technology and problems in the real applications.The method of SCNN is to constructive neurons inside the network for mapping sub-set data dynamically, but not the fixed structure, which is suitable for external training data and its further changes. This training has the character which is scalable and could be suitable for further data. From the experiences compared in the paper, we give the characters of SCNN. Meanwhile, the paper discussed the multi-dimensional data abilities, and also the real model. The main research context is established and discussed in the below: Firstly, this paper discussed the comparison of the theories and methods between SCNN and classical Neural Network. The constructive method and study produce is described and gives the basic principle. Meanwhile, from the experiences, this paper gives the rate comparison between the SCNN and full connection Neural Network. SCNN has more high speed of constructive neural network with the some error recognition rate, which is more suitable for great mount analysis. All of these give the theory basis for further research.Secondly, this paper gives the discrete and feasible constructive method based the discussion of common constructive produces. From the experiences, these methods are been verified effectively, and also give the effective on high-dimensional data analysis. Further more, SCNN is not to find the optimize hyper-plane in high dimension space, which is N-P problem, but to use the neural network to make the data description. For this mapping produce, this method is more suitable for data frequency, time compactable, multi-dimensional space. Comparing the classical neural network, SCNN is more effective for multi-dimensional method analysis.Combining with the goodness, multi-dimension data analysis is proposed by this paper. With the different side of SCNNs, the description is described with SCNNs, which gives the data analysis by data set. On the other side, this paper gives the pre-process method based on SCNN ,the decomposition of data set, and also the multi-dimensional data analysis.Lastly, on the background of cheapness-rent-house in our country, we apply the rent-exponent based on the research of SCNN. From the different section of SCNN described the data set, we give the exponent number of rate house. The system is valuable and has the real application fields.From the valuable research work above, this paper gives the multi-dimension data analysis and the reason of solving the weakness of classical neural network. Application research is applies in the real rent house. SCNN has the further goodness in the high-dimensional data with the new method. In the last paper, it discusses the related work and gives the further work with abbreviated depiction.