Design and Implementation of Artificial Neural Network Classifiers based on Hypertuning Parameters for Breast Cancer Diagnosis

Abstract

Breast cancer is an extremely prevalent disease which impacts women all over the world, and minimizing the death rate among women requires detection at an early stage. This research employs Artificial Neural Network (ANN) to predict and detect breast cancer at an early stage. The suggested model is trained and tested using the 569 patient records taken from Wisconsin Diagnostic dataset for breast cancer. The dataset is split between training and testing sets using 80:20 ratio, with the random states being 0, 1, and 2 for each set. With random state 1 benign and malignant cases are split in same ratio as were in original dataset. The binary cross-entropy function is used to calculate the discrepancy between the predicted class probability and the actual class label. In this research, an Adam optimizer is employed to reduce the loss function. Overall, the parameters of the model are tuned to produce the best outcomes for particular tasks depending on the choice of different random states, activation function, loss function, and optimizer. Several measurements, including training and testing accuracy, confusion matrix, precision, sensitivity, F1_Score, FPR, MCC (Matthews correlation coefficient), AUC, and ROC curve, are employed to evaluate the model performance. Our experimental findings show that the suggested model is capable of making precise predictions and diagnosis of breast cancer.