International Conference on Computer and Knowledge Engineering

Home / 11th International Conference on Computer and Knowledge Engineering

A Graph-based Feature Selection using Class-Feature Association Map (CFAM)

Authors :

Motahare Akhavan¹ Seyed Mohammad Hossein Hasheminejad²

1- Department of Computer Engineering, Faculty of Engineering, Alzahra University 2- Department of Computer Engineering, Faculty of Engineering, Alzahra University

Keywords :

Feature Selection, CFAM, Community Detection, Graph Analysis

Abstract :

The quality and size of the feature space play an essential role in the main process of machine learning and directly affect the results of data mining. Feature selection as a preprocessing step in data mining is used to find the effective features while eliminating redundant and irrelevant ones from the initial set of features. One way to understand the relationships among the features is to represent them as a graph. In this paper, a graph-based method is proposed for feature selection. To represent the feature space as a graph, we use the Class-Feature Association Map (CFAM) concept, which considers both relevance and redundancy and is constructed using Spearman's correlation measure. Then we apply the Leiden algorithm for detecting communities of the graph and finally use eigenvector centrality and variance score to find the most significant features from the communities. According to experimental results, our algorithm reduces an average of 70.6% of features while achieving higher or competitive classification accuracy with the selected features.