HOW COULD WE MAKE OUR ALGORITHM BETTER?

While we're happy with the accuracy of our current model, there are many ways in which it could be improved. For instance, we had a limited amount of training data (189 normal and 38 malignant), but more images might have helped our algorithm achieve higher accuracy. We also used only test 15 images of both types (normal and malignant), and more test images would have given us a better sense of the success of our code. Since we're tailoring our features to a small set of test images, we may be overfitting to a very specific group of images. 

​

The accuracy of our model could also be improved, both from the preprocessing and the feature extraction standpoints. As we've learned from this project, machine learning can feel random -- it's not always easy to understand why a certain algorithm works well. Doing even deeper research into how specific combinations of preprocessing and feature extraction produce high accuracy could better inform our model. 

​

In addition, we have limited computational time and power, so we can't test all the combinations of preprocessing, feature extraction, and classification. We could find ways to make our code more efficient, reducing computation time so that we could better test combinations.