I have collected some images of around 20 Active Regions and separated them into training and test sets to run a CNN solar flare prediction model. This project was originally inspired by several resources, but the main paper which I have replicated is mentioned in the repository. There are two models which show drastically different results.
- The paper itself inspires first model.
- To improve the accuracy, I used a pre-trained model from the TensorFlow hub.
I will not name the models here as I want you to open the notebook and look at the results and understand that sometimes sticking to something blindly is not so good :P