For all its simplicity, expectation-maximization performs incredibly well in a range of scenarios. That being said, there are some potential limitations that we need to be aware of:
- Expectation-maximization does not guarantee that we will find the globally best solution.
- We must know the number of desired clusters beforehand.
- The decision boundaries of the algorithms are linear.
- The algorithm is slow for large datasets.
Let's quickly discuss these potential caveats in a little more detail.