|
this is regarding comparing logistic regression with neural networks.while finalizing the logisitc regression model it happnes that we have to reduce the variables and keep only the significant ones. but in neural networks this may not be the case. there is generally not anything like picking up the significant variables while fixing the model. So in this case is reducing the logisitc model and comparing it with neural networks a fair comparison or we should keep the logisitc regression as it is without reducing the variables. |
|
For the best possible comparison, you should compare your logistic regression with a subset of variables to the following three things: 1. neural network that doesn't select the most important variables 2. neural network trained only on the variables selected by the logistic regression 3. neural network that uses the first layer weights to select the important variables In the case of (3) you should be able to do something analogous to what you do in the logistic regression case for selecting variables. Thanks George. Due to time constraint I won't be able to do according to what you have said, but surely consider in the coming task. I have at hand logistic regression on say 10 variables(these variables are as they are in the original set without applying any significance criteria). And i have neural networks results on 10 variables. With these 2 results in place what can you say about them? Is it okay to say that no variables reduction scheme is applied and this is how the classification algorithms performs on 10 variables. Please give me a supporting answer to the analysis carried out.
(Nov 24 '11 at 02:04)
Dataminer21
|
