Detection of Francisella Tularensis Pathogen in Soil using Neural Networks

Francisella Tularensis is a dangerous bacterium that can cause diseases in both humans as well as animals. The detection of Francisella Tularensis in soil is vital to prevent widespread epidemic of Tularemia disease. Classification of soil samples based on its characteristics can be helpful in initial detection of this pathogen. In this paper, we study the detection of Francisella Tularensis in soil using backpropagation neural networks. As number of neurons and hidden layers along with activation and loss function play an important role in the performance of networks, different experiments were conducted to study their effects. It is concluded that a backpropagation network having one hidden layer with ten neurons performs best for our dataset when activation function is tanh and loss function is absolute. The network with these configurations gives an accuracy of 82.61 % for ten-fold cross validation.