Unbalanced Data Set Classification Based on Convolutional Neural Network

Abstract

When a data set contains different categories of data and the number of elements contained in the categories differs greatly, we call this type of data set an unbalanced data set. When classifying imbalanced data sets, it is difficult to correctly classify important minority data using traditional algorithms. How to better classify minority data has become a difficult point. The use of convolutional neural networks can solve this point well. Therefore, the purpose of this article is to study the classification of unbalanced data sets based on convolutional neural networks. This article first analyzes the structure and algorithm of the convolutional neural network. Combining with the current research status of imbalanced data set classification, the difficulties and performance evaluation methods are discussed. Finally, the classification of imbalanced data sets is researched and analyzed with convolutional neural network. This paper systematically expounds the classification model of convolutional neural network, the computational complexity of convolutional neural network, and the analysis of parameters and models of convolutional neural network applied to the classification of imbalanced data sets. And use observation method, comparative method and other research methods to carry out experimental investigation on the research of this article. Experimental research shows that the CNN algorithm has better precision and recall results for different data set samples in the classification process than the SVM algorithm and the K-means algorithm, and the classification is more accurate.