Build and train a computer vision model to detect malaria from images of infected red blood cells. Model uses a CNN neural network to classify parasitized and uninfected cells with a 98.69% accuracy. EDA was done to identify key characteristics in the data that then informed the solution strategy. It is proposed that this is a potential replacement for traditional testing, with supporting cost benefit analysis and implementation strategy.
Multiple architectures where tested including:
- Varying loss functions
- Varying levels of coplexity
- Varying activation functions
- data augmentation
- transfer learning From there the most best model was hyper parameterized to optimize performance.
Link to Dataset: https://drive.google.com/file/d/1n3o1Xghpy9ufZwHkQFE5l5d9sUHQOUWM/view?usp=sharing