Cuckoo Search-Based Optimization for Cancer Classification: A New Hybrid Approach

Area of Science:

• Bioinformatics and Machine Learning
• Computational Biology
• Genomics and Cancer Research

Background:

• Predicting cancer from high-dimensional, imbalanced microarray data presents significant challenges in bioinformatics and machine learning.
• Existing dimensionality reduction techniques offer varied performance with different classifiers and datasets, necessitating optimized feature selection.
• Independent Component Analysis (ICA) effectively extracts independent components suitable for Naïve Bayes (NB) classification criteria.

Purpose of the Study:

• To develop an optimal framework for cancer prediction from complex microarray datasets.
• To address the optimization challenges in selecting features extracted by ICA for improved NB classification.
• To introduce and evaluate a novel hybrid metaheuristic algorithm for feature selection in high-dimensional biomedical data.

Main Methods:

• Feature extraction using Independent Component Analysis (ICA) on microarray data.
• Development and implementation of a hybrid cuckoo search (CS) and artificial bee colony (ABC) algorithm (CS-ABC) for optimizing ICA-extracted gene subsets.
• Classification of microarray data using the Naïve Bayes (NB) classifier with features selected by the CS-ABC-ICA hybrid method.

Main Results:

• The novel CS-ABC algorithm, integrated with ICA, was successfully applied for the first time to address dimensionality reduction in high-dimensional microarray data.
• The hybrid CS-ABC algorithm demonstrated improved local search capabilities, enhancing the optimization of gene subsets.
• The CS-ABC-ICA approach significantly improved the classification accuracy of the Naïve Bayes classifier compared to previous feature selection methods.

Conclusions:

• The proposed CS-ABC hybrid algorithm effectively optimizes feature selection for ICA-extracted genes, leading to superior classification performance.
• This method provides a robust solution for dimensionality reduction and enhances the accuracy of cancer prediction from imbalanced microarray data.
• The CS-ABC-ICA framework offers a promising approach to overcome premature convergence and achieve better results in complex biomedical data analysis.