Feature selection in gene expression data using principal component analysis and rough set theory

Area of Science:

• Data Mining
• Machine Learning
• Pattern Recognition
• Signal Processing

Background:

• High-dimensional data with numerous features is common in data mining and machine learning.
• Feature selection is crucial for preprocessing such data to improve classification tasks.
• Traditional methods include Feature Extraction (FE) and Feature Selection (FS).

Purpose of the Study:

• To develop an effective feature selection method for high-dimensional data.
• To combine Principal Component Analysis (PCA) with Rough Set Theory for improved feature selection.
• To enhance classification accuracy by selecting the most adequate principal components.

Main Methods:

• Principal Component Analysis (PCA): An unsupervised linear FE method for dimensionality reduction by identifying directions of maximal variance.
• Rough Set Theory: A method for discovering data dependencies and reducing attributes using data alone.
• Rough PCA: A hybrid approach combining PCA for feature extraction and Rough Set Theory for feature selection from principal components.

Main Results:

• The proposed Rough PCA method successfully identifies principal features from high-dimensional datasets.
• Upper and Lower Approximations from Rough Set Theory were applied to find a reduced set of features.
• The method was validated on gene expression data, demonstrating its effectiveness.

Conclusions:

• Rough PCA offers a robust approach for feature selection in high-dimensional data.
• The joint application of PCA and Rough Set Theory ensures selected features are highly discriminative for classification.
• The method shows promise for applications in bioinformatics, particularly with gene expression data analysis.