5

We have used multi-linear regression (stepwise) approach to perform spatial modeling with parameters derived from Remote Sensing data. The outcome of the statistically significant components produce respectable RMSE with acceptable R Square and adjusted R Square. However, it is notable that the identified significant parameters violate VIF and Tolerance range in multi-collinearity diagnostics. The produced model is closed to reality with very good RMSE value.

In this backdrop, can we use this model by ignoring VIF and tolerance statistics because, in spatial domain, co-relation between variables is found to be a common phenomenon.

Improve this question

1 Answer 1

Reset to default
2

I think you must review a multicollinearity before you uses model regression. It is part of assumptions of multiple Linear Regression. And if you dont accomplish with all parameter, you have a bad model or false model. If you want to ignore VIF and tolerance statistics, maybe you can uses a no-parametric model or if you have a big date, you can uses machine learning, Random forest or anorther that ignored. But I think so you can reduce the multi-collinearity effect, if you evaluated your variables. Bye :)

Improve this answer
4
  • Yes, it is quite convincing. We might not deviate from the fundamentals.
    – Rex
    Sep 29, 2017 at 9:14
  • If you have a big date, you can uses machine learning. Or uses no-parammetrics model
    – Dinosca
    Sep 29, 2017 at 12:39
  • Yes, I would like to switch over. would you please let me know what other machine learning techniques might be conducive as a replacement of stepwise regression?
    – Rex
    Oct 2, 2017 at 3:37
  • 1
    Depend that, what size have your data? if you have a big date, maybe you could uses Random Forest regression (randomForest library) or neuronal network. But in the other hand, if you want to use linear regression, you could uses regsubset. This specific function reduce a data, and try to remove multicollinearity
    – Dinosca
    Oct 2, 2017 at 13:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.