Figure out if your professor is actually not capable of advising this project, or if you are just not communicating well. For instance, oftentimes when I hear that advisors are giving inconsistent advice, it just happens that the student often misunderstands what the advisor is trying to say (or the advisor is just throwing out ideas in a brainstorming style, which almost by definition means that some of those ideas will be conflicting - that does not necessarily mean that he does not know what he is talking about). For instance:
it's a machine learning project, and while the first critical thing is training data, he told me to focus on finding features.
Focusing on the features first does not seem entirely crazy to me. Of course you'll need to find good training data for supervised learning at some point, but identifying the potential features is indeed usually a very early step in the process, afaik.
sometimes he want it to be unsupervised, and when I mention training data he switched to supervised system.
It's a research project, he probably does not have a 100% clear vision of what exactly you need to do to make it work (nor should he). Maybe he is just throwing out ideas, and you are taking his words too seriously?
Critically question whether he or you are not understanding the project correctly. If this is the "main project he gets funding from", chances are that actually you are not understanding correctly.
If you decide that he actually cannot advise the project, and you can't do it alone, you can either get help or get out.
Getting help may include finding a co-advisor, but this may be difficult for various reasons (one being that most faculty are not overly interested in joining in ongoing, troubled projects of other professors for the dubious honor of being a co-advisor). Usually, it is easier to get the help of other persons in your lab more senior than you (postdocs, or even older PhD students), who have the necessary skills, and collaborate with them on your project. For this to work, you need to figure out what is "in it" for your collaborator (joined papers? something useful for their own dissertation / projects?).
If you are unable to get help, I see no other way than to get out, that is, change your project.