Analysis of Various Optimizer on CNN model in the Application of Pneumonia Detection

Pneumonia is a possibly lethal bacterial respiratory infection caused by bacteria, fungi, or viruses that invade the alveoli of the human lung with a load of fluid or pus. The most often used technique for diagnosing pneumonia is by examining Chest X-rays, and the results of the X-ray must be evaluated by a medical specialist. The difficult way of identifying pneumonia results in the loss of life due to insufficient diagnosis and treatment. With the advent of computer technology, the development of an automated device for detecting and treating pneumonia is now feasible. In this study, different Optimizers were used on a different number of layers to find which combination is the most efficient for the purpose of Pneumonia Detection using CNN. The Convolutional Neural Network models were designed involving one, two, three, four and five hidden layers for classification purposes and three optimisers namely RMSProp, Adam and SGD were chosen. For training and testing purposes, a Kaggle dataset consisting of 5856 Chest X-Ray images was used, split into three folders of the test, train and val. The model with four hidden layers and the "SGD" optimiser achieved the highest testing accuracy of 91% and the model with one hidden layer and the "Adam" optimiser achieved the lowest testing accuracy of 84%.