Universal Feature Selection Tool

Tool Repository

---

User Manual | JavaDoc

The Universal Feature Selection Tool (UniFeat) is an open-source tool developed entirely in Java for performing feature selection process in different areas of research. The project aims to create a unified framework for researchers applying feature selection.

UniFeat provides a set of well-known and state-of-the-art feature selection methods within the essential auxiliary tools, including performance evaluation criteria, visual displays, statistical analysis, and reduced datasets to compare the performance of feature selection methods.

Citation

Sina Tabakhi and Parham Moradi, Universal Feature Selection Tool (UniFeat): An Open-Source Tool for Dimensionality Reduction, Neurocomputing, Vol. 535, pp. 156-165, 2023.

If you find this repository useful, please cite our paper:

@article{tabakhi2023unifeat,
 title     = {Universal feature selection tool (UniFeat): An open-source tool for dimensionality reduction},
 author    = {Tabakhi, Sina and Moradi, Parham},
 journal   = {Neurocomputing},
 year      = {2023},
 volume    = {535},
 pages     = {156-165},
 publisher = {Elsevier},
 doi       = {10.1016/j.neucom.2023.03.037}
}

Objectives

Our aim in the development of UniFeat as a comprehensive feature selection tool includes six aspects.

1. UniFeat implements well-known and state-of-the-art feature selection methods within a unified framework.
2. UniFeat can be considered a benchmark tool due to the development of methods in all feature selection approaches.
3. The functions presented in UniFeat provide essential auxiliary tools needed for performance evaluation, results visualization, and statistical analysis.
4. UniFeat has been implemented completely in Java, and can therefore be run on a wide range of platforms.
5. Researchers are able to use UniFeat through its GUI environment or as a library in their Java codes.
6. Finally, the open-source nature of UniFeat can help researchers use and modify the tool to fit their research requirements and greatly facilitate it for them to share their methods with the scientific community rapidly.

Algorithms supported

In UniFeat, you can simply access to the well-known and state-of-the-art feature selection methods in the literature. The project has three tabs corresponding to the different feature selection approaches, including Filter, Wrapper, and Embedded tabs. In the Filter, Wrapper, and Embedded tabs, the UniFeat repository has involved the feature selection methods, the details of which are listed in Tables 1-3, respectively.

Table 1: Filter-based feature selection methods in the UniFeat repository.

method	supervised/unsupervised	multivariate/univariate
Information gain	supervised	univariate
Gain ratio	supervised	univariate
Symmetrical uncertainty	supervised	univariate
Fisher score	supervised	univariate
Gini index	supervised	univariate
mRMR	supervised	multivariate
Laplacian score	supervised & unsupervised	univariate
Relevance-redundancy feature selection (RRFS)	supervised & unsupervised	multivariate
Term variance	unsupervised	univariate
Mutual correlation	unsupervised	multivariate
Random subspace method (RSM)	unsupervised	multivariate
UFSACO	unsupervised	multivariate
RRFSACO_1	unsupervised	multivariate
RRFSACO_2	unsupervised	multivariate
IRRFSACO_1	unsupervised	multivariate
IRRFSACO_2	unsupervised	multivariate
MGSACO	unsupervised	multivariate

Table 2: Wrapper-based feature selection methods in the UniFeat repository.

method	supervised/unsupervised
Binary particle swarm optimization (BPSO)	supervised
Continuous particle swarm optimization (CPSO)	supervised
Particle swarm optimization version 4-2 (PSO(4-2))	supervised
HPSO-LS	supervised
Simple GA	supervised
HGAFS	supervised
Optimal ACO	supervised

Table 3: Embedded-based feature selection methods in the UniFeat repository.

method	supervised/unsupervised
Decision tree based method	supervised
Random forest	supervised
SVM_RFE	supervised
MSVM_RFE	supervised
OVO_SVM_RFE	supervised
OVA_SVM_RFE	supervised

More information is available at the UniFeat Homepage.

License

UniFeat is implemented and developed principally at the University of Kurdistan, Sanandaj, Iran, and distributed under the MIT License terms.