Best Geometry Feature Selection Of Neurons Categories

The brain is the vital body’s organs, composed of so many neurons in different strctures and features. Currently, neurons classify according to the geometry is essentially in brain science. But researches on neuronal geometry feature selection method is less, and the complexity and exactitude degree of classification is affected by feature selection significantly, so this article will do some try in this filed.In this essay, we try to determine the significant neuronal geometry features for classification through feature extraction, to reduce the complexity and improve the exactitude degree of classification. In samples of sensory neurons and motor neurons spatial form data, we use several different feature selection and extraction methods, namely to determine the optimal feature subset, then use each of these feature subset and same classification model to classify neurons. Based on classification results, we comparative analysis various features of selection and extraction method, from which we select the category of best feature subset, and get the best one.First, the matrix neurons data in disparate dimensions which reflecting spatial form are converted to 43 geometry features, then according to the sensory neurons and motor neurons spatial form chart compares the difference between the two, visualize feature data, a rough analysis of the impact categories of major feature. Secondly, feature select by three dimensionality reduction method like feature selection and feature extraction. First, using information gain ratio(IGR) based on the principle of decision tree and branch and bound(BBA)based on class separabily criterion get the optimal feature subset. Then, feature extraction by PCA respectively, for the original features and features subsets got by BBA method. Finally, in order to evaluate classification performance of each method, the neurons were classified on the selected feature using decision trees and SVM.The results show that, classification performance are good, whatever classifier of the three methods of selecting the features. Wherein the feature selection method based on IGR selected characteristic of the linear correlation is weak, and the classification is the best, but the dimension reduction is slightly inferior than the BBA method. The features selected by BBA associated with strong linear correlation, so the effect is slightly inferior to the classification based on the IGR feature selection method. The classification performance of subset of features got by PCA is slightly less than the other two methods’. Among them, the performance characteristics of classification based on PCA of the original features is superior than PCA of features subset got by BBA. Although the classification performance of the subset, choose by PCA of features subset got by BBA, is the worst, but it is the best dimension reduction. Balanced complexity and the classification performance, PCA of features subset got by BBA is the best choice for data capacity, high dimensionality feature many big data classification.This method can be applied to even the whole area of the brain neurons of big data and multi-class classification in order to study the function of neurons in the biomedical foundation.