I take a constructive step when working on a problem. Be it programming or my daily doings. This helps me learn a little faster and gives me a clearer understanding of what I'm to do next after each successful step.
Learning ML is no doubt complicated but I hope with this problem solving flywheel that you too can discover as well.
It's important you start by defining your problems. What problems are you trying to solve? Is it a supervised learning problem or unsupervised learning problem?
What kind of data do we have? Is it structured (eg. CSVs, spreadsheets, databases etc) or unstructured (Images, Audio etc.)
Define what success means for that project. What Accuracy, precision or recall scores determines the success for your model. E.g. 95%, 80% etc.
What do you already know about the data? Apart from modelling this data, what have you learnt about this data you are working with. This is mostly where Data visualization plays a great role.
Based on your problem definition and data, what machine learning model should we use?
Now we have modelled this data, has the success we earlier defined been achieved. No? What can we do better? What can we try next? Should we try another machine learning model? Should we try tuning the parameters of our current model?
This is how I try solving each problem I face as a machine learning Intern. What's missing? Help me add it in the comment section. 😊