Maybe you can do a dual-use project, where, if the ML part turns out to be not that useful, at least the data you collected or the baseline model you developed

Maybe you can do a dual-use project, where, if the ML part turns out to be not that useful, at least the data you collected or the baseline model you developed have a positive impact as well.