Metaheuristic integrated machine learning classification of colon cancer using STFT LASSO and EHO feature extraction...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Microarray gene expression data presents a high-dimensional challenge, often leading to overfitting and reduced classification accuracy.
• Effective dimensionality reduction is crucial for enhancing the interpretability and performance of predictive models in genomics.

Purpose of the Study:

• To investigate the efficacy of Short Term Fourier Transform (STFT), Least Absolute Shrinkage and Selection Operator (LASSO), and Elephant Herding Optimisation (EHO) for feature extraction in lung cancer microarray data.
• To enhance lung cancer classification accuracy and interpretability through optimized feature selection and dimensionality reduction.

Main Methods:

• Applied STFT, LASSO, and EHO for feature extraction from lung cancer microarray datasets.
• Utilized various classifiers including Gaussian Mixture Model (GMM), Particle Swarm Optimization (PSO)-GMM, Detrended Fluctuation Analysis (DFA), Naive Bayes classifier (NBC), Firefly-GMM, Support Vector Machine with Radial Basis Kernel (SVM-RBF), and Flower Pollination Optimization (FPO)-GMM for classification.

Main Results:

• The combination of EHO feature extraction and FPO-GMM classification achieved the highest accuracy (96.77%), with an F1 score of 97.5, MCC of 0.92, and Kappa of 0.92.
• Demonstrated the effectiveness of STFT, LASSO, and EHO in reducing data dimensionality and identifying significant features.

Conclusions:

• The proposed STFT, LASSO, and EHO methodologies significantly improve feature extraction and dimensionality reduction for microarray gene expression data.
• These advanced techniques contribute to more accurate and interpretable early diagnosis of lung cancer.