Evaluation of Activation Functions in CNN Model for Detection of Malaria Parasite using Blood Smear Images

Activation functions are an essential parameter in deep learning models, primarily when it is deployed with CNN. These deep neural networks not only deal with linear data but handles different non-linear classifications as well. Automatic malaria parasite detection using blood smear images is a popular classification application by using CNN architecture. Several models use various activation functions for this parasite detection. Choosing among these activation functions for a custom model sometimes is a time-consuming task. In this paper, we have evaluated Sigmoid, Hyperbolic Tangent Function (Tanh), Rectified Linear Units (ReLU), Leaky ReLU, and Swish activation functions for malaria Parasite detection in the CNN model. It has been observed in our empirical results that the Swish function performs better than others in terms of accuracy and loss value on a given malaria parasite images dataset.