I'm trying to automatically identify peaks using scipy's find_peak function. Currently, I'm attempting to optimize the prominence parameter so that only "real" peaks are identified without capturing too many noisy points. Everything is normalized and min-max scaled, so the range of my data is 0-1. I've been experimenting with prominence=0.01, but this results in peaks in my data (shown below) not being identified. If I try to lower the prominence to prominence<0.01, then I pick up on a lot of noise. I was wondering how I can fix this, considering the peaks that aren't being identified seem quite prominent. Image of two overlayed graphs with noisy points labeled as peaks on one graph but no labeled peaks on the other graph: Graph with prominent peaks but no identified peaks using find_peaks: Graph with many noisy points labeled as peaks: Continue reading...
