Leveraging the Effect of CNN to Identify Severity of Lumpy Skin Disease in Cattle

There have been research articles in the past that identified the presence of Lumpy Skin Disease (LSD) in Cattle through machine-learning methods. This paper is focused not only on confirming the presence of the virus but also on predicting the stage of the LSD virus in a Cattle skin image sample. In this paper, several machine-learning classification algorithms including Support Vector Machines, Naive Bayes, Random Forest, AdaBoost, K-nearest neighbors, and finally, Convolutional Neural Networks, were analyzed to classify the image samples of Cattle skin into 3 categories: Normal (LSD-negative), Mild LSD, and Severe LSD classes. A custom-made dataset consisting of 3454 image samples from the three classes was built to train the models. There is no cure for LSD, and the infected cattle's immune response is the only way out; however, as a supporting aid, primary anti-viral treatment can be given for a faster recovery. This should be administered to the cattle with a higher probability of getting cured and staying alive. Hence, the detection of the stage of the disease is significant in terms of the investment of time and treatment to needy cattle as healthcare facilities and government resources are limited. After analyzing the results from the models trained on the algorithms mentioned earlier, the best accuracy was obtained through Convolutional Neural Networks with pre-trained base model VGG-16 with 92.93% on the training dataset and 88.93% on the validation dataset. This model eliminates possibilities of human bias to empower the government in granting fair compensation in case of death due to LSD and to maximize resource utilization in the best possible way if integrated into a user-friendly application.