Can we use KNN for predicting continuous target variable?

knn_reg
machine_learning

#1

Hi, i’m trying different algorithms available for prediction. Can i use KNN for predicting continuous variable?.


#2

Hi @Niteshdm
difficult question as the answer is driven by the current density of existing observations. First keep in mind continuous could mean different properties. So in business we never deal with continuous variables or responses if we refer to mathematical properties.
So yes if you are in a narrow space knn could give you you could predictions (remember based on past history) now if you are in a space with many dimensions (lot of X) for example 100 or more thousands as it could be seen in business … no knn will not solved your problem ( you have no density the famous curse of dimensionality ) or you have one huge amount of points. Then dimension reduction should be used, such as Lasso if you go will linear type of model ( I am bias here ).
In few words … give more details for more direct answer this help. Next keep in mind density and distance.
Hope this help.
Alain


#3

Hi, @Lesaffrea
Thanks for helping me understand better.