aabo5/Malaria-Detection ? reverse-engineered prompt

Build me a malaria cell image classification project that can take microscope blood smear images and tell whether each cell is Parasitized or Uninfected.

I want the full workflow set up, from loading the NIH dataset and cleaning it, to resizing the images, normalizing them, splitting them into train, validation, and test sets, and saving the prepared data so it can be reused. Then train and compare six models, a custom CNN plus VGG16, ResNet50, MobileNetV2, EfficientNetB0, and DenseNet121, using the same data and evaluation setup so the results are fair.

Please include training plots, accuracy and F1 scores, a clear comparison table, and Grad CAM heatmaps so I can see what the models are looking at. If it helps, organize it with notebooks and a small set of helper scripts, and make it easy to rerun the whole pipeline from start to finish. If you need to check current docs or best practices for any model, go ahead and look them up online.