Machine Learning with Swift

Research the existing approaches

The key question here is, how do other people solve similar problems?

Can you use native iOS SDK to solve your problem? For example, if you want to detect people's faces in photos, you don't need to train your own neural network or Haar cascade. Just use the Vision framework instead. In other words, do not reinvent the wheel. Look for ready solutions that can work on the device or server side. For the most common daily tasks, you will find something appropriate.

Expanding the range of searches, perform a literature review. Even if you've not found ready solutions, you at least will get useful insights into the approaches and the domain specifics. The sites that will be handy at this stage are arXiv, Google ...