Pneumonia, a lung disease that affects both younger children and older adults, causing respiratory problems. Pneumonia disease has been known to cause several deaths, and there is a need for a faster method of diagnosing the disease for early treatment. The use of machine learning techniques in various fields has delivered excellent results as it has been applied to address challenges encountered in society. It has been utilized in several domains like medical, environment, security, industrial, business, finance and many more. Machine learning applications are characterized by the ability to process massive amounts of data, identify patterns, relationships, and provide logical interpretations in the identification, classification or prediction of an outcome. One aspect of machine learning called transfer learning utilizes deep convolution. In this research, the transfer learning technique is applied to classify pneumonia. The radiologist's readings could be subject to inter-class variability, and delayed results can occur. The methodology involved the application of five transfer learning techniques, namely, MobileNet-V2, GoogLeNet, AlexNet, DenseNet, and VGG-19. The individual experimental results show an accuracy of 0.79, 0.96, 0.96, 0.83, and 0.91, respectively. The results revealed that GoogLeNet and AlexNet have the best accuracy, while VGG-19 followed closely with 0.91, DenseNet and MobileNet-V2 record the lowest. GoogLeNet and AlexNet have a high precision of 0.99. The weighted ensemble applied gives an approximate accuracy of 0.97. This research shows an improvement on the other models reviewed and can be deployed to fast-track the detection of pneumonia.

Pneumonia detection classification transfer learning deep learning ensemble