Study on an improved Lie Group machine learning-based classification algorithm

Abstract

In order to address the problem that the geodesic distance of Lie Group machine learning method has failed to use the sample class information, this paper proposes an improved Lie Group machine learning algorithm. Firstly, this algorithm maps the vector space consisting of sample dataset into a differential manifold, and each sample corresponds to a point in the manifold structure; then, the differentiation value is used to replace the Euclidean distance for calculation of the geodesic distance between samples, and the class of test sample is determined according to the geodesic distance relation between the test sample and the sample in training dataset. The results of classification experiment prove that compared with the traditional Lie Group machine learning classification (LGC) algorithm, the K-nearest neighbor (KNN) algorithm and support vector machine (SVM) classification\algorithm, the improved classification algorithm has higher classification accuracy.