Classification Method Research Based On Neuron Space Form

The human brain plans aims to establish common standards worldwide database of multidisciplinary integrated analysis of large amounts of data to speed up the brain's awareness. The results of the "brain Plan" will help the human thorough understanding of the brain's way of operation,to further clarify the occurrence of consciousness, thinking process and a series of puzzles, also for the treatment of Parkinson's disease, Alzheimer's disease and other brain diseases lay a solid foundation. China is also actively promoting the Chinese characteristics of the "China Brain Project." Neurons are the basic unit of the brain structure, which is the basis and importance of its research. It is an important topic to distinguish different types of neurons according to the spatial pattern, but the problem is still not well resolved.This article from the external visual characteristics of neurons- space form to start using now the mainstream pattern recognition method to classify neurons. Regional cerebral cortex neurons were collected from the database of the most common four types of neurons data: interneuron, GABAergic, nitrergic, pyramidal.. Because data is stored in SWC format, so the first neuronal morphology attribute space calculation, the 43 property characteristics. Then feature selection, selected 22 features. After several pattern recognition method using feature selection and classification simultaneously- decision trees, Bayesian networks, artificial neural networks, support vector machine classification of neurons, the final results of the analysis.The results showed that three methods are ideal. Classification tree best, followed by artificial neural networks, support vector machines followed, somewhat less Bayesian network. Decision Tree accuracy was 99.55%. Easy to understand and explain the decision tree, each step of the calculation and decision rules has a corresponding logical expression.But because of the sample size of neurons in each category, may cause deviations in computational modeling category. Neural network accuracy rate of 99.22%. Its parallel distributed processing capability, can fully approximate complex nonlinear relationship.But it takes a lot of parameters, different parameter settings affect the classification results.Support vector machine accuracy rate of 99.116%, which handles high-dimensional data with good results, But the improper selection of kernel function leads to the poorclassification performance of support vector machines. Bayesian network accuracy rate of90.38%. Which is based on probability theory, causal relationship between variables with.The causal relationship between the variables of this article is not very clear, so the result is not satisfactory classification.The study of these classifications can be more different types of neurons more accurate classification, as well as to predict the unknown neurons lay the foundation for functional studies, are available for medical and other related fields.