How to identify critical parameters in a dataset?
Say that you have a dataset of 10 different parameters (listed in 10 columns) to describe a phenomenon (listed in the 11 column). What would be a procedure to identify which of the 10 parameters are critical to predicting the phenomenon on hand (i.e. maybe 4-5 parameters are need to properly predict the phenomenon, how to go around figuring out what are these parameters?).
Thanks,
Find more posts tagged with
The respective models are used for this. While for some of those models like linear regression (a formula) or a decision tree (nested if-then statements) a mathematical representations can be derived, this is not general the case. Especially for the more complex models (and therefore often also more accurate models) like GBT or Neural Nets there is a close to zero chance to transform them into a human-readable format. That is the reason why we have tools like the Simulator or Explain Predictions in the first place - so that you can build trust in what the models are doing even without such an explicit formula.
Best,
Ingo
Auto model will provide you with attributes (Parameters) that support the prediction. If you don't have access to automodel, you can also see the correlation of these attributes to output using the correlation matrix operator. You can look at some feature selection technique operators like optimize selection, forward selection
If you are trying to predict using different algorithms like a decision tree, SVM, etc., you can use explain predictions operator to see which attributes are helpful in predictions for a particular algorithm
Thanks
The respective models are used for this. While for some of those models like linear regression (a formula) or a decision tree (nested if-then statements) a mathematical representations can be derived, this is not general the case. Especially for the more complex models (and therefore often also more accurate models) like GBT or Neural Nets there is a close to zero chance to transform them into a human-readable format. That is the reason why we have tools like the Simulator or Explain Predictions in the first place - so that you can build trust in what the models are doing even without such an explicit formula.
Best,
Ingo
Auto model will provide you with attributes (Parameters) that support the prediction. If you don't have access to automodel, you can also see the correlation of these attributes to output using the correlation matrix operator. You can look at some feature selection technique operators like optimize selection, forward selection
If you are trying to predict using different algorithms like a decision tree, SVM, etc., you can use explain predictions operator to see which attributes are helpful in predictions for a particular algorithm
Thanks