I have a spectrogram of a signal and want to classify it. I used different features such as MFFC, the density of power spectrum and other common features and classify using the random forest, but I did not get good results.
I think that my features are not well selected and I guess that there is no relation between my aimed classes and the selected features. my classes are so rare.
could anyone suggest a way for obtaining features?(for very rare data) also, I used some methods of feature extraction from biological images, but I think that the data are different. my data are double and those are some 3 dimensions data.