What does the OOB estimate of error in a RandomForest model imply?




I built two randomForest models on my training dataset which gave me comparable accuracy in the output but the OOB estimate of error was almost half for one than compared to the other. What does it imply about the two models? What is the difference? Is the one with lower OOB error estimate better?



you did not mentioned but I guess the accuracy you refer to is the one made on the training set. The OOB should be interpreted as one measure done on a test test (similar to cross validation). With Random Forest you are in ensemble therefore the final accuracy will be better than the one on the OOB.
I guess here that first you speak first of the accuracy on the training set and the internal OOB used by random forest.