Breast Cancer Detection Based on Modified Harris Hawks Optimization and Extreme Learning Machine Embedded with Feature Weighting

• Published in: Neural processing letters Vol. 55; no. 4; pp. 3631 - 3654
• Main Authors: Jiang, Feng, Zhu, Qiannan, Tian, Tianhai
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.08.2023; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Artificial neural networks; Breast cancer; Classification; Complex Systems; Computational Intelligence; Computer Science; Datasets; Diagnosis; Exploitation; Feature selection; Machine learning; Medical diagnosis; Model accuracy; Optimization algorithms; Particle swarm optimization; Sensitivity; Support vector machines; Weighting; Breast cancer detection; Extreme learning machine; Particle swarm optimization; Feature weighting; Harris hawks optimization
• ISSN: 1370-4621; 1573-773X
• DOI: 10.1007/s11063-021-10700-w

Computer-aided diagnosis (CAD) can assist doctors with clinical diagnosis and improve diagnosis accuracy and efficiency further. It is significative and valuable for cancer detection by using machine learning. In this paper, a hybrid model based on optimization algorithm and machine learning with feature weighting is carried out to detect breast cancer. Firstly, to surmount the limitation of nonlinear and imbalanced data distribution, we apply feature weighting (FW) based on K-Means to make benign and malignant samples more separate. Then Particle Swarm Optimization (PSO) is used to enhance searching ability of Harris Hawks Optimization (HHO). Moreover, the HHO optimized by PSO (PHHO) is employed to optimize Extreme Learning Machine (ELM). Finally, in order to verify the availability of our proposed FW-PHHO-ELM model, experiments are implemented on Wisconsin Diagnosis Breast Cancer (WDBC) data set. The results indicate that the proposed model can achieve 98.76%, 97.37% and 99.46% on accuracy, sensitivity and specificity respectively. The comparison results demonstrate that the proposed model outperformed reported benchmark models and existing models on accuracy and sensitivity. Besides, the proposed model could also balance sensitivity and specificity well.