Statistical approaches such as frequency ratio, statistical index and decision decision-making approaches like AHP have been used and they are still popular.

How do you separate machine learning from statistical approaches? They are basically somewhat similar right?

Some papers considered the methods you described also as machine learning.

It is true that some ML use statistical techniques for learning, but the statistical approaches that I mentioned comprise much simpler calculations for specifying weights to the sub-classes of landslide conditioning factors. For instance frequency ratio uses a simple division. You can refer to some of my works to see details.

Similarly, other statistical techniques such as Information Value, Shannon's entropy, Weight of Evidence, etc. can also be used for a fairly reliable estimate of landslide susceptibility, in addition to the aforementioned ones. All of these techniques include basic mathematical calculations.