Malaria Cell Infection Detection

For a university assignment I was tasked with creating and comparing different models on a malaria cell infection dataset. It was a binary classification problem with an existing dataset of fairly simple data.

I decided to use a range of models, many of which were pulled from existing packages as this was not an assignment of creating models, but comparing their characteristics and performance on a given dataset for a given problem.

The models chosen were, Inception, a Multi-Layer Perceptron, MobileNetV3 , ResNet50, VGG and EfficientNet. After training, I attempted to continue training on both the VGG and EfficientNet models to see if there were any remarkable improvements, shown on the graphs with the "Long" prefix.

All models performed quite well on such a basic task, with the exceptions of MobileNetV3 (as this is a lower accuracy, faster real-time model), and the plain Multi-Layer Perceptron, both of which were not particularly surprising but I thought were good for comparison's sake.

The custom models used were both convolutional models built using TensorFlow's Keras package.