I'm working on a project where I have replicates of time series data and I'm attempting to identify and remove outliers. I've done a manual review of the data and you can see clear outliers that should be flagged and I'm trying to determine the best and most reproducible way to do this. For my initial attempt I structured the data in a Pandas DataFrame where the rows are the samples and the columns are the time points. I tried z-scoring but with this data all z-scores are < 3 so no outliers are detected even though there is a clear difference in one of the wells. One challenge I'm having is that these replicate numbers are usually low (3-4) and the z-scoring method worked well when I had higher replicates but not as well with these lower replicate numbers. Are there other ways I should be thinking about outlier detection when working with multiple time series data? Is there a method that would work well at low n?
pivot.apply(zscore).apply(np.abs)
Picture of DataFrame after Z-scoring
Simple picture of time series replicates showing clear outlier
