Breast Cancer Detection Using Optimization-Based Feature Pruning and Classification Algorithms

Early and accurate detection of breast cancer reduces the mortality rate of breast cancer patients. Decision-making systems based on machine learning and intelligent techniques help to detect lesions and distinguish between benign and malignant tumours.
In this diagnostic study, a computerized simulation study is presented for breast cancer detection. A metaheuristic optimization algorithm inspired by the bubble-net hunting strategy of humpback whales is employed to select and weight the most effective features, extracted from microscopic breast cytology images, and optimize a support vector machine classifier. Breast cancer dataset from UCI repository was utilized to assess the proposed method. Different validation techniques and statistical hypothesis tests (t-test and ANOVA) were used to confirm the classification results.
The accuracy, precision, and sensitivity metrics of the models were computed and compared. Based on the results, the integrated system with a radial basis function kernel was able to extract the fewest features and result in the most accuracy (98.82%). According to the tests, in comparison with genetic algorithm (GA) and particle swarm optimization (PSO), the WOA based system selected fewer features and yielded higher classification accuracy and speed. The statistical validation of the results further showed that this system outperformed the GA and PSO in some metrics. Moreover, the comparison of the proposed classification system with other successful systems indicated the former’s competitiveness.
The proposed classification model had superior performance metrics, less run time in simulation, and better convergence behaviour owing to its enhanced optimization capacity. Use of this model is a promising approach to develop a reliable automatic detection system.