I am working on a project that involves classifying four diseases. In my dataset, I have two types of data:
- Type-1: This contains information related to duration or time intervals (tabular data).
- Type-2: Signal data.
Using Type-1 data, I have extracted statistical features such as mean, SD, etc., achieving an accuracy of 86% with a Keras classifier. The model takes only a few minutes to train.
For Type-2 data, I perform wavelet analysis to obtain image data from the signals. I use the GoogLeNet model, which yields an accuracy of 96%. However, the training of this model takes 150 minutes.
My question is, as similar work has been done in the literature, what unique approach can I take to add novelty to my work? Can anyone provide ideas for brainstorming? Your input will be greatly appreciated.