基于相关信息熵和粒子群算法的特征选择方法

Feature Selection Based on Correlation Information Entropy and Particle Swarm Optimization

为了在特征选择中获得具有较高分类准确率的特征子集,考虑特征之间、特征与类别的相关性对于其类间区分能力的影响,提出了一种基于相关信息熵的特征排序消除法和二进制粒子群算法相结合的特征选择算法。本算法先采用相关信息熵特征排序算法,对数据集进行特征重要度排序,选取SVM作为分类器,快速去掉部分无关特征,初步缩减数据纬度,然后以改进的粒子群算法继续搜索最优子集,将排序算法得到的优良子作为粒子群算法的部分初始种群,使后续粒子群算法有一个较好的搜索起点。实验结果表明,采用该方法筛选出来的特征自己具有更好的分类性能,并且有效降低了系统的检测时间。

To obtain a subset of features with higher classification accuracy,consider the influence of the correlation between the features and the classes or features on the classification,a feature selection method based on correlation information entropy and Binary Particle Swarm Optimization algorithm was proposed.Firstly,this method filtrated irrelevant features to reduce the feature dimension respectively by feature sorting based on correlation information entropy,then continued to search optimal subset,and used some better subset of feature sorting based on correlation information entropy as part of initial binary particle swarm optimization population with good starting points.The simulation test results showed that the better classification performance was obtained according to the selected optimal feature subset,and the testing time of the system was reduced effectively.