In simple words: The most complex model is not automatically the best model.
Start simple
A rule-based system follows conditions written by people. For example: if an assignment is missing, send a reminder. This needs no training data and is easy to explain.
When learning helps
Machine Learning (ML) can find patterns from enough reliable past examples. For structured questionnaire answers, a conventional ML model may be a sensible first model. Deep Learning (DL) is often useful for complex text, images, audio, and video, but it may need more data and computing power. Compare methods on new data before choosing.
See the flow
A guidance website can begin with transparent study-habit rules. With consent and trustworthy outcome data, it may later test an ML model.
Remember: The most complex model is not automatically the best model.