Genetic algorithm based cancerous gene identification from microarray data using ensemble of filter methods.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Microarray datasets are vital for cancer detection but suffer from high dimensionality.
• Irrelevant and redundant features in microarray data pose significant classification challenges.
• Feature selection is essential to enhance accuracy by removing uninformative genes.

Purpose of the Study:

• To propose a novel 2-stage feature selection model for microarray datasets.
• To address the NP-hard problem of optimal feature subset selection using meta-heuristic techniques.
• To develop a classifier-independent model for robust cancer detection.

Main Methods:

• A 2-stage approach combining an ensemble of filter methods (ReliefF, chi-square, symmetrical uncertainty) with a genetic algorithm (GA).
• Ensemble filters utilize the union and intersection of top-ranked features.
• GA is applied to refine feature subsets for improved selection accuracy.

Main Results:

• The proposed 2-stage model demonstrated superior performance compared to existing methods.
• The union of features in the ensemble filter stage outperformed the intersection.
• The model proved classifier-independent, tested with Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), and K-Nearest Neighbor (K-NN).

Conclusions:

• The developed 2-stage feature selection model effectively handles high-dimensional microarray data for cancer detection.
• The ensemble filter and GA approach provides a robust method for identifying optimal gene subsets.
• This approach offers improved accuracy and classifier independence in cancer classification tasks.