SLSP: Sparse Learning and Similarity Preserving

Official implementation of the IEEE paper:

Similarity Preserving Unsupervised Feature Selection based on Sparse Learning

Published in:

2020 10th International Symposium on Telecommunications (IST 2020), IEEE

Authors:

• Hadi Zare
• Mohsen Ghassemi Parsa
• Mehdi Ghatee
• Sasan H. Alizadeh

---

Keywords

Unsupervised Feature Selection, Similarity Preserving Learning, Sparse Learning, Feature Selection, Clustering, Representation Learning, Spectral Analysis, Dimensionality Reduction, Data Mining, Machine Learning.

---

If You Use This Code, Please Cite the Paper

This repository accompanies the following IEEE publication.

@inproceedings{Zare2020SLSP,
  author={Zare, Hadi and Parsa, Mohsen Ghassemi and Ghatee, Mehdi and Alizadeh, Sasan H.},
  title={Similarity Preserving Unsupervised Feature Selection based on Sparse Learning},
  booktitle={2020 10th International Symposium on Telecommunications (IST)},
  year={2020},
  pages={50--55},
  doi={10.1109/IST50524.2020.9345884}
}

DOI:

https://doi.org/10.1109/IST50524.2020.9345884

⭐ If this repository helps your research, please consider starring the repository and citing the paper.

---

Overview

SLSP is an unsupervised feature selection framework that simultaneously preserves both global and local similarity structures in data while identifying the most informative features.

The proposed method integrates:

• Global similarity preservation
• Local similarity preservation
• Sparse learning
• Clustering-guided feature selection

into a unified optimization framework.

Unlike many conventional feature selection methods that only preserve local neighborhood information, SLSP jointly exploits global and local data structures to improve clustering performance.

---

Main Contributions

✔ Global similarity preservation using Symmetric Nonnegative Matrix Factorization (SymNMF)

✔ Local similarity preservation through spectral graph analysis

✔ Sparse feature learning using ℓ₂,₁-norm regularization

✔ Joint optimization of clustering and feature selection

✔ Convergence-guaranteed iterative algorithm

✔ Extensive evaluation on benchmark datasets

---

Why SLSP?

Many existing unsupervised feature selection methods focus on only one aspect of data structure.

Method	Local Similarity	Global Similarity	Clustering	Joint Learning
MCFS	✓	✗	✗	✗
UDFS	✓	✗	✗	✓
NDFS	✓	✗	✓	✓
SPFS	✗	✓	✗	✓
GLSPFS	✓	✓	✗	✓
SLSP	✓	✓	✓	✓

SLSP combines all three components within a unified framework.

---

Experimental Results

The proposed method was evaluated on ten benchmark datasets:

Dataset	Domain
ALLAML	Bioinformatics
Colon	Bioinformatics
GLIOMA	Bioinformatics
Lung	Bioinformatics
BA	Image Analysis
COIL20	Object Recognition
ORL	Face Recognition
Yale	Face Recognition
Isolet	Speech Recognition
Madelon	Artificial Dataset

---

Performance Highlights

SLSP was compared against:

• LS
• MCFS
• UDFS
• NDFS
• GLSPFS
• SPUFS

Key findings reported in the paper:

• Best clustering accuracy on six benchmark datasets.
• Consistently ranked among the top-performing methods.
• Significant improvements in NMI on ALLAML and Colon datasets.
• Robust performance across biological, image, speech, and artificial datasets.

---

Evaluation metrics:

• Accuracy (ACC)
• Normalized Mutual Information (NMI)

Each experiment was repeated 20 times and the mean and standard deviation were reported.

---

Applications

SLSP can be applied to:

• Bioinformatics
• Gene Expression Analysis
• Cancer Classification
• Computer Vision
• Face Recognition
• Speech Processing
• Pattern Recognition
• Data Mining
• Representation Learning
• Clustering Preprocessing

---

Citation

Please cite the following paper if you use this repository:

@inproceedings{Zare2020SLSP,
  author={Zare, Hadi and Parsa, Mohsen Ghassemi and Ghatee, Mehdi and Alizadeh, Sasan H.},
  title={Similarity Preserving Unsupervised Feature Selection based on Sparse Learning},
  booktitle={2020 10th International Symposium on Telecommunications (IST)},
  year={2020},
  pages={50--55},
  doi={10.1109/IST50524.2020.9345884}
}

---

License

This repository is provided for academic and research purposes.

---

Contact

Mohsen Ghassemi Parsa

Email: mgparsa@ut.ac.ir GitHub: https://github.com/mohsengh

For questions, suggestions, bug reports, or collaborations, please open an issue.

---

Support the Project

If you find this work useful:

⭐ Star the repository

📄 Cite the paper

🔄 Share the repository

These actions help increase the visibility and impact of the research.