Choosing a learning style is mostly a question of what feedback is available. If every example comes with a correct answer, supervised learning applies directly. If there are no answers but the data has internal structure worth exposing, unsupervised learning fits. If the only signal is a reward received after acting, reinforcement learning is the natural choice.
The diagram below presents three problem descriptions. For each one, the learner selects the style whose feedback signal matches the problem, and the diagram reveals whether the choice is consistent with the available signal. The reasoning to apply is: what is the feedback, and when does it arrive? A labeled answer available at training time points to supervised learning. No answer at all, but a desire to group or compress, points to unsupervised learning. A reward that arrives only after a sequence of actions points to reinforcement learning.