05-15-2016 01:22 AM
In Regression Model , it is possible that a predictor may have positive relation with dependent variable in Bi-variate analysis , however in Model Equation the predictor may have a negative sign. Why this happens and how is this resolved.
This can happen in Trainin & Validation data as well. That is a predictor may have a Positive sign in the Model equation in Training data set and negative sign in Validation Data-set.
05-15-2016 02:51 AM
To me, this indicates a problem somewhere, either in selection of the model, or partitioning of the data into training and validation sets.
Do you have an example that you can post?