<p>Oncogene diagnosis is a critical field that utilizes oncogene data for disease prognosis and informing treatment strategies. This study introduces an attribute reduction algorithm that combines attribute relevance with neighborhood rough set and variable precision rough set theories, and is carefully optimized for use with nine oncogene diagnosis datasets. To evaluate the effectiveness and performance of our algorithm, we conducted rigorous comparative assessments with a range of machine learning and deep learning algorithms. Additionally, we proposed a dynamic ensemble classifier model based on cooperative game theory, specifically designed for tumor classification and diagnosis. This model leverages the simplified data to improve diagnostic accuracy. Extensive experimental evaluations were conducted, comparing the performance of various algorithms across different datasets. The results highlight the significant advantages and promising applications of our proposed approach in tumor genetic diagnosis. Overall, this study opens a new pathway for tumor genetic diagnosis, demonstrating the ability to enhance diagnostic accuracy and efficiency, and shows that the proposed attribute reduction algorithm has certain advantages over existing rough set-based reduction methods. Comparative analyses with current machine learning and deep learning techniques indicate that our approach exhibits robustness and potential in advancing oncogene diagnosis.</p>