Multiclass Prediction of Pneumonia based on X-Rays by Using Mining Techniques

Recently, the entire world has suffered from a novel virus known as “COVID-19” because of which millions of people have been hospitalized, and more deaths have been caused. Health organizations and doctors use the Polymerase Chain Reaction (PCR) test, a standard test to identify infection in the human body but identifying the victims is very time-consuming, which takes nearly two to three days. So, developing an AI-based system to scan the chest image through X-rays or CT scans will enable a viable solution. This research study employs a deep learning model to classify patient conditions into normal, pneumonia, and covid. Because of the scarcity of data, the dataset is increased by combining image manipulation and augmentation techniques. The model could extract the features from chest X-rays by using autoencoders and decoders. Finally, the proposed model implements transfer learning by retraining the weights and customizing the last layers of CNN to make it a multiclassifier instead of a binary classifier. Here, 96% of accuracy is achieved for the proposed model, and the loss is negligible. Compared to its counterpart models, the proposed model has achieved a better accuracy.