From my interviewing experience, I have seen candidates are most comfortable with this step (especially candidates with not a lot of industry experience). Having said that, a lot of candidates answer in a biased manner regarding which model they would pick, rather than focussing on the current problem. It’s important to know the algorithms that you are discussing thoroughly as the interviewer is evaluating you on your in-depth knowledge as well as modelling choices.

Here are the things you should keep in your mind when talking about modelling approaches:

• Modelling choices and training process. 
• Technical details of the training method and contrast between different approaches (trade-off analysis).

Positive signal in this area usually includes candidates showcasing familiarity with a variety of modelling techniques and selecting the one appropriate in this context, justifying the selection and doing trade-off analysis. The candidate should give technical details of the different algorithms listed and discuss various potential issues that could arise with these approaches. It’s important to realise that we are trying to implement a solution in a large-scale setting, so walking through the challenges and how to mitigate them shows experience and gives a very positive signal.