I am working on python using Random Forest classifier to predict unknown contaminate location and I have got 88% prediction accuracy but most of the error come from miss-classification of nearby contaminate locations. How can I make the top 5% miss-classification errors to be considered as true when it calculates the accuracy?
As shown in the graph, more than 60% of the error on predicting contaminate location32 comes from location31. How can I specify wrong but nearby predictions from the real location, as a true prediction?
I wouldn't try to change that because you want answers as accurate as they can be. Try to tune the algorithm, basically you can do that with scikit-learn or in R with caret.
In most of the cases an algorithm like XGBoost will outperform randomforest and you may get a better but not biased result.
- Badges
- Science topic
- Similar topics
- Books