Yes, it is a good idea to use machine learning for stuttering detection. I would recommend you start with how to extract mfcc features in MATLAB, refer this link for the same:
These mfcc features are representations of your audio file, they are a 2D matrix similar to an image. Image recognition algorithms such as convolution can be then be used for stuttering recognition, where the output can be a stuttering label.
You might also find these links useful:
