engleski

Dataset Size-Based Approach In Design Of Artificial Neural Network For Breast Cancer Diagnosis

One of the challenges in medical data classi- fication is the determination of the sufficient training dataset size. In this research, the ef- fect of training dataset size on various arti- ficial neural network (ANN) configurations is shown. All ANNs are trained and tested using Wisconsin Breast Cancer (Diagnostic) Dataset, which contains 569 samples. Dataset is divided into training dataset of 410 samples and test dataset of 159 samples. The training procedure for all ANNs is performed with datasets in the range between 10 and 410 samples. Performances of all ANNs are eval-uated using ROC analysis. From obtained results, it can be seen that if datasets smaller than 282 samples are used for training of ANN, higher AUC values will be achieved if deep ANN designed with ReLU activation function is used. On the other hand, if larger datasets are used, the best performances will be achieved if ANNs are designed with one hidden layer and Logistic sigmoid activa-tion function. Our results showed, that it is possible to use smaller datasets for classifier training if right ANN architecture is utilized. Furthermore, it can be concluded that there is no need for using large datasets for design-ing an ANN for breast cancer diagnosis. (PDF) Dataset Size-Based Approach in Design of Artificial Neural Network for Breast Cancer Diagnosis. Available from: https://www.researchgate.net/publication/344152 374_Dataset_Size- Based_Approach_in_Design_of_Artificial_Neural_N etwork_for_Breast_Cancer_Diagnosis [accessed Sep 07 2020].

Activation function ; Artificial neural network ; Breast cancer diagnosis ; Dataset size

nije evidentirano