.png){kind=icon}
.png){kind=icon}
Return
Graph classification based on graph set reconstruction and graph kernel feature reduction
.png){kind=icon}
OA
AI
DOI：10.1016/j.neucom.2018.03.029.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Graph, a kind of structured data, is widely used to model complex relationships among objects, and has been used in various of scientific and engineering fields, such as bioinformatics, network intrusion detection, social network, etc. Building an automatic and highly accurate classification method for graphs becomes quite necessary for predicting unknown graphs or understanding complex structures among different categories. The kernel method is regarded as a powerful solution to graph classification, which consists of two steps, namely, graph kernel mapping and classification. However, the feature selection process is ignored, and those sub-structures with low discriminative power result in classification accuracy decrease. In order to solve this problem, we propose an efficient graph classification algorithm based on graph set reconstruction and graph kernel feature reduction. First of all, the least discriminative frequent subgraphs and part of the infrequent subgraphs are removed to reconstruct the original graph set. Then we adopt the graph-kernel-based discriminant analysis method to perform feature reduction on the well-reconstructed graph set. At last, the whole framework of the graph classification method is introduced and any commonly used classifiers can be utilized. Extensive experimental results on a series of bioinformatics benchmarks show that our graph classification algorithm demonstrates a significant improvement of prediction comparing with other graph-kernel-based classification approaches. (C) 2018 Elsevier B.V. All rights reserved.
Keywords:
Graph classification
Discriminative subgraph
Graph kernel
Kernel discriminant analysis
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
Organization
No organization information available