Novel HGDBO: A Hybrid Genetic and Dung Beetle Optimization Algorithm for Microarray Gene Selection and Efficient Cancer Classification

Introduction: Ovarian cancer ranks as the seventh most frequently diagnosed cancer and stands as the eighth leading cause of cancer-related mortality among women globally. Early detection significantly improves survival rates and outcomes, highlighting the need for enhanced screening methods and increased awareness to facilitate early diagnosis and treatment. Microarray gene data, characterized by its high dimensionality, includes the expression levels of thousands of genes across numerous samples, posing both opportunities and challenges in the analysis of gene functions and disease mechanisms. Method: This paper presents a novel hybrid gene feature selection method called HGDBO, which combines the Dung Beetle Optimization (DBO) algorithm with the Genetic Algorithm (GA) to increase the effectiveness of microarray data analysis. The proposed HGDBO method utilizes the exploratory capabilities of DBO and the exploitative strengths of GA to identify the most relevant genes for disease classification. Experimental results on multiple microarray datasets demonstrate that the hybrid approach offers superior classification performance, stability, and computational efficiency compared to traditional and state-of-the-art methods. To classify ovarian cancer, Naïve-Bayes (NB) and Random-Forest (RF) classification algorithms were employed. Results and Discussion: The proposed Random Forest model outperforms the Naive Bayes model across all metrics, achieving better accuracy (0.96 vs. 0.91), precision (0.95 vs. 0.91), recall (0.97 vs. 0.90), F-1 score (0.95 vs. 0.91), and specificity (0.97 vs. 0.86). Conclusion: These results underscore the effectiveness of the HGDBO method and the Random Forest classifier in enhancing the analysis and classification of ovarian cancer using microarray gene data.