poor performance on other data

[afadda1]

we're using human data from different sources/labs to try and predict the protocol used. We used the already trained model on human data. the predictions were erroneous: e.g. for ATAC dataset of 48, only 4 were predicted correctly. for WES data of 6, 2 were predicted as ChIPseq
We don't know if it would be useful to retrain the model on the data we have to get better results. what do you recommend?
thanks