An AUC-based multi-kernel weighted support vector machine ensemble algorithm for breast cancer diagnosis

Statistical Theory and Related Fields, vol. 10, pp. 251–267

Abstract

Machine learning algorithms have demonstrated outstanding performance for disease diagnosis. Kernel function selection plays a crucial role in effectively transforming the nonlinear nature of input data. To enhance breast cancer diagnosis, we propose a novel ensemble algorithm, namely, AUC-Ada-L1MKL-WSVM, which integrates Weighted Support Vector Machines (WSVM), AdaBoost, and Multi-Kernel Learning (MKL). This ensemble algorithm introduces two main innovations. First, it simultaneously updates the weights of training samples and the combined kernel function during classification. Second, it incorporates an AUC-based approach to adjust training sample weights, effectively controlling the growth rate of misclassified sample weights in AdaBoost. Experimental results are provided to demonstrate the effectiveness of our method, which achieves an AUC score of 97.21% and an accuracy of 97.64% on the WDBC dataset, and an AUC of 97.53% and an accuracy of 97.46% on the WBC dataset. Comparative analysis further confirms that our ensemble algorithm outperforms four benchmark models in classification accuracy.

Authors 4

1. Mushuang Cheng

   Shanghai University of Engineering Science

   Affiliation as printed

   Shanghai University of Engineering Science

2. Lintong Liu

   Shanghai University of Engineering Science

   Affiliation as printed

   Shanghai University of Engineering Science

3. Haixiang Lin Aachen

   Leiden University · Delft University of Technology

   Affiliation as printed

   Delft University of Technology

   Leiden University

4. Guoqiang Wang corresponding

   Shanghai University of Engineering Science

   Affiliation as printed

   Shanghai University of Engineering Science