During last two decades, breast cancer, especially for females, has been a major health concern around the globe. The fact that early and efficient detection of it may reduce the mortality rate necessitates the development of efficient diagnostic instruments or process. In the proposed work, an artificial neural network (ANN)-based methodology is adopted to work on the Wisconsin Breast Cancer (WBC) dataset obtained from Kaggle. It includes a total of 570 sample data with 30 features each. To train the network, the number of neurons in hidden layer is changed and the accuracy of the output is observed. The network parameters are saved against the number of neurons in the hidden layer providing maximum accuracy. A comparative study on how the number of nodes in the hidden layer can affect accuracy is also conducted. Through rigorous training, testing, and validation, an accuracy rate of 99% has been achieved. Additionally, the most significant characteristics in the diagnosis of breast cancer are also identified. The factors such as compactness_worst, fractal dimension_se, symmetry worst, smoothness_se, and symmetry_se play a crucial role in distinguishing benign and malignant tumors. This study contributes to the prevailing efforts of the research community to improve the fast but accurate detection of breast cancer by the potential use of ANN in medical domain and offering analytical information about the significance of features. The proposed work may lead to early and accurate identification of breast tumor and thereby starting of the treatment towards complete recovery from breast cancer.