I am looking for an existing R code that could do band selection using hyperspectral images. I have seen codes that do random forest for instance, but I hoping to use a code that does an ensemble of results from various models based on the importance of the bands.

Apart from RF, some of the models for band selection include PLS, SVM, KNN. Is there a code that integrates all of them (and maybe more models) into one?


The only method that you mentioned that meets the criteria of being an ensemble method and providing variable importance is random forests. I would direct you to the parameter selection methods available in the rf.modelSel function in the rfUtilites package.

If by "ensemble" you mean, multiple modeling approaches, please abandon this idea. This imposes something called the meta-model effect where different models assume different error structures. This makes it very difficult to evaluate the cumulative performance and one can inadvertently incorporate error distributions that vary wildly.

| improve this answer | |

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