Collaboration Graph for Feature Set Partitioning in Data Classification
The collaboration graph project is about a new method for feature partitioning in data classification. In machine learning, data can be divided into different partitions for classification and different features can be extracted from the data as input for machine learning models. However, having a large number of features can make classifiers suffer, which is known as the curse of dimensionality, so dividing features into smaller sets is one approach to solving this problem.
In this project, a collaboration graph is used for feature partitioning. In this method, each node is assigned to a feature and each edge represents the collaboration relationship between two features. By applying the community detection method on the collaboration graph, the features are partitioned. Then, a classifier is trained for each partition and the final classification result is obtained by fusing the classifiers’ results of each partition.
Using this method, the issue of the curse of dimensionality is solved, the accuracy and speed of the classification process are improved. In this project, several different data sets, including high-dimensional real and synthetic data, have been used to train and test this method. The results show that the use of collaboration graph can improve the accuracy and speed of the classification process and solve the curse of dimensionality problem.
In the following, we will examine this project further. In the introduction section, the need to reduce the dimension of features in data classification has been investigated and various methods have been introduced for this task. Then, the collaboration graph is introduced for feature partitioning and how they work is explained. Also, in the sections of related works, other methods for partitioning features have been introduced and compared with the proposed method.
In the proposed method section, more details are given about the collaboration graph for feature partitioning. In the experiment section, the results have been reported on different datasets. The results show that the use of collaboration graph can improve the accuracy and speed of the classification process and solve the curse of dimensionality issue.
In the conclusion section, the advantages of using the collaboration graph for feature partitioning in data classification have been investigated. Also, suggestions for future work in this field have been presented.
Members: Khalil Taheri, Hadi Moradi, and Mostafa Tavassolipour
Related Papers: Khalil Taheri, Hadi Moradi, and Mostafa Tavassolipour. "Collaboration graph for feature set partitioning in data classification." Expert Systems with Applications 213 (2023): 118988.