基于AdaBoost的神经元形态分类的研究

Research on Classification of Neuron Morphology Based on AdaBoost

根据神经元形态的几何特征,使用AdaBoost算法对其进行分类,采用决策树、贝叶斯和关联规则分类模型作为基分类器。算法首先采用直接面向组合分类器分类精度提升的集成学习算法选取基分类器,其次利用分类过程中生成样本的累计权值来调整前K次(K>1)被错误分类样本的权重,并提出双重阈值法对样本的最终投票表决结果进行判定。对20个测试样本进行分类,得出高可信度分类数为18个。

According to the geometric characteristics of neuron morphology, neurons were classified by AdaBoost algorithm, in which the decision tree, Bayesian and association rules classification model were used as base classifiers Firstly, the improvement of the classification precision of combined classifier directly oriented ensemble learning algorithms were adopted to select the base classifier. Secondly, cumulative weights of the sample generated during the classification were used to adjust the weights of misclassified samples in the former K （K〉 1） classification processes, and the dual threshold values for final voting results of samples were set. The 20 test samples were classified by AadBoost algorithm, and the results show that the number of highly reliable classification is 18.