| Huge quantities of data are collected from a diverse set of services in E-commerce.Various data mining technologies have been developed to meet the need of speed,the large volumes of data in the age of big data.Instances selection aims to select informative instances and remove redundant/noisy instances from the original training dataset,which has played a crucial role in many practical applications.The instance selection methods reduce the volume of data used in learning models and speed up the training process.Thus instance selection is essential for data preprocessing.Support Vector Machine(SVM)is a widely-used method in machine learning community due to its sound theoretical foundation and good performance.Firstly,a brief literature review is given on instance selection.Then an instance selection method,NES,is proposed,which selects instances based on the information of nearest enemy neighbor.The whole original dataset is divided into several independent subsets and the instances in every subset have the same nearest enemy neighbor.Instance selection can be executed simultaneously in these subsets.In addition,a new concept of ?-Net is defined and a novel instance selection method,NENet,is proposed based on ?-Net accordingly to select instances near the decision boundary.NENet can select valuable instances not only on the convex surface but also on the concave surface.Experimental results show that the two proposed methods outperform other conventional instance selection methods. |
