Optimization of artificial neural network structure and hyperparameters in hybrid model by genetic algorithm: iOS–android application for breast cancer diagnosis/prediction

• Published in: The Journal of supercomputing Vol. 80; no. 4; pp. 4533 - 4553
• Main Author: Bülbül, Mehmet Akif
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.03.2024; Springer Nature B.V
• Subjects: Accuracy; Artificial intelligence; Artificial neural networks; Back propagation; Back propagation networks; Breast cancer; Compilers; Computer Science; Datasets; Decision support systems; Decision trees; Deep learning; Diagnosis; Fuzzy logic; Genetic algorithms; Hybrid structures; Interpreters; Machine learning; Mammography; Medical diagnosis; Neural networks; Optimization algorithms; Processor Architectures; Programming Languages; Success; Support vector machines; Decision support systems; Model and hyperparameter optimization; Breast cancer classification; Genetic algorithm; Artificial neural networks
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-023-05635-z

Breast cancer is a common disease that can result in death among women. Cancer research is important because early detection of cancer facilitates clinical practice for patients. The aim of the study is to ensure that breast cancer can be diagnosed in a short time and easily. For this purpose, a dataset containing 116 samples, 9 features and 2 target variables (Breast Cancer Coimbra) from the UCI library was used during the training and testing phases. A hybrid structure was created with genetic algorithm (GA) and artificial neural network (ANN) to classify the datasets. With the established hybrid model, the feedforward backpropagation artificial neural network model and the hyperparameters in this model structure have been optimized with the genetic algorithm. The performance of the structure constructed with the most successful gene parameters obtained was compared with weighted K-nearest neighbors, decision tree, and linear support vector machine methods. In all machine learning methods used, fivefold cross-validation was applied and the dataset was divided into two groups as 50% training and 50% testing in order to test the models with different data. The hybrid model proposed in the study performed better than other machine learning methods with 100% correct classification rate. Although there are few data in this study, the accuracy is higher than other literature. In addition, an iOS–android-based application has been developed for the diagnosis and prediction of the disease with the findings obtained. Thanks to the developed application, the most important factor in the fight against the disease, time and cost spent for the diagnosis of this disease will be saved. Considering the interest in artificial intelligence techniques in cancer research, this study presents a new diagnostic method and a usable application in terms of patient decision support systems.