I have used LIME on tabular data of ECG signals which have 34 features, and my aim to to figure out which features are deemed important for the respective classes. I used submodular pick with a sample size of 100 and number of features as 5. The explanations that I am receiving are confusing me further as the features are changing with every explanation. So if there are local explanations for the class 'Interruption', there are a few features that are consistently appearing such as Heart rate, but I don't know how I can summarise my findings. Should I make a frequency table where I write the number of times a feature is appearing in an explanation? Any input would be much appreciated!
I have used LIME on tabular data of ECG signals which have 34 features, and my aim to to figure out which features are deemed important for the respective classes. I used submodular pick with a sample size of 100 and number of features as 5. The explanations that I am receiving are confusing me further as the features are changing with every explanation. So if there are local explanations for the class 'Interruption', there are a few features that are consistently appearing such as Heart rate, but I don't know how I can summarise my findings. Should I make a frequency table where I write the number of times a feature is appearing in an explanation? Any input would be much appreciated!