My standard response to such questions is: Read papers. Use their datasets. A dataset must be citable to be reanalyzed from public sources, and when people publish data then there is a paper associated, otherwise the dataset would not be out. Also, you anyway must know the literature around the topic to focus on novelties and not just replicate what others have analyzed before. If you find an interesting paper, check for data availability, check the methods sections whether the setup is what you need. Check the papers to know what already has been analyzed, and then try to find novelties.