Traditional classification algorithms struggle with the high dimensionality of medical data, resulting in reduced performance in tasks like disease diagnosis. Feature selection (FS) has emerged as a crucial preprocess...
详细信息
Traditional classification algorithms struggle with the high dimensionality of medical data, resulting in reduced performance in tasks like disease diagnosis. Feature selection (FS) has emerged as a crucial preprocessing step to mitigate these challenges by extracting relevant features and improving classification accuracy. This paper proposes a hybrid FS method, FJMIBCOA, which integrates Fuzzy Joint Mutual Information (FJMI) as a filter measure and Binary cheetah optimizer algorithm (BCOA) as a wrapper method. Unlike existing hybrid FS methods, the proposed method employs FJMI to address uncertainty in feature relationships, providing several advantages such as handling both discrete and continuous features, accommodating linear and non-linear relationships, noise robustness and effectively utilizing intra- and inter-class information. It also employs BCOA as a wrapper method, requiring a few parameters, minimizing computational overhead and enhancing classification robustness, making it an efficient and adaptable solution for FS in complex medical datasets. The proposed method is validated on 23 medical datasets and 14 high-dimensional microarray datasets, demonstrating excellent performance in terms of fitness value, accuracy and feature size. FJMIBCOA surpasses existing methods in medical datasets by achieving higher accuracy in 78.26% of datasets while reducing the feature size by 84.79%. Similarly, in microarray datasets, it improves accuracy in 78.58% of datasets with an impressive 95.08% reduction in feature size. Furthermore, FJMIBCOA achieves superior accuracy in 60% of datasets while selecting fewer features in 78.57% of datasets as compared to previous studies. Statistical testing indicates that FJMIBCOA outperforms other methods significantly. The proposed method enhances diagnosis accuracy and minimizes medical testing requirements, making it suitable for real-world, high-dimensional datasets and decision-making in medical data analysis. The findings
暂无评论