基于蚁群算法的行为识别特征优选方法

为了降低特征冗余,提高移动用户行为识别的准确率,提出一种基于蚁群算法的移动用户行为识别加速度信号特征优选方法。首先对样本数据进行预处理,根据特征对不同行为的分类敏感度进行初次优选,降低特征搜索空间的维度;然后利用蚁群算法结合神经网络分类器,以特征的分类准确度为评价准则对特征集合进行了二次优选。实验结果表明,该方法优选出的特征集具有较好的识别性能。

A feature selection method for behavior recognition based on ant colony algorithm

To reduce redundancy features and improve accuracy of mobile user behavior recognition, an acceleration signal feature selection method is proposed for mobile user behavior recognition based on ant colony algorithm. The sample data is preprocessed and features are optimized initially according to the classification sensitivity of every feature of the different behavior to reduce dimension of the feature search space. Then, ant colony algorithm combined with neural network classifier is used to make a secondary accuracy for evaluation criteria. Experiment results show method has a better recognition performance. optimization and feature classification that the feature set selected by the