Hyper Parameter Optimization of Convolutional Neural Networks for Breast Cancer Classification
K. Sreekala, Jayakrushna Sahoo · 2021
The Convolutional Neural Networks (CNN) have recently become the most effective way of classifying images. Numerous researchers have demonstrated the value of neural network design to improve efficiency by optimizing the hyperparameters of the neural network. Some researchers have shown the importance of the neuron’s activation by using various activation functions and tuning numerous parameters in deep neural networks. This study presents efficiently classified breast cancer images as Benign or Malignant by using deep learning schemes. The proposed scheme uses the pre-processing techniques for images such as normalization and standardization, and performs the classification by using convolutional neural network. Also hyper-parameter tuning of convolutional neural network is performed by using two efficient optimization techniques as grid search and randomized search technique. The experiment is conducted by using the INbreast dataset. The results demonstrated reveals that the hyper-parameter tuning/optimization techniques along with normalization and standardization techniques is the most effective technique for the mammogram image classification with CNN to classify breast cancer significantly.