In every research project I have been a part of or conducted, I have always wanted more data, even if we thought we raised enough birds to get enough samples.The conclusion of the study mentions one of the other points that I see as a big problem, the small sample size:
"In this study, a total of 60 used eggs, and only 47 of them were hatched. This number is comparatively low, and a large number of data is needed for better results."
And I also see a problem with them training the computer on those 47 chicks, then having it predict the sex of those SAME chicks to check the accuracy. So even with all the right answers, the computer had a pretty low accuracy (better than random chance, worse than the usual after-hatch sexing methods.)
"37 out of 47 (0.787) chicks were classified correctly."
(Quotes are from the previously linked study:
https://pmc.ncbi.nlm.nih.gov/articles/PMC9832119/)
I'd like to see further research on the matter, to prove out whether it really does or does not work. My bet is that it won't work, but proving it one way or the other would be better than forever wondering. I don't personally care enough to do controlled experiments.
Less than 50 eggs hatched is a pretty small population to start with.
For my honors thesis, we set the same amount of quail eggs for 4 different bloodlines (a lot of eggs per line, I forgot how many, hatch rate wasn't a part of the study.) One of the lines didn't hatch well, and we almost had to modify the sample size based on that line because we only had so many surviving birds when it was time to draw blood.