I am training a classifier using python random forest classifier. Here is detail of my data:
total training data= 7656
class 1=1052
class 2= 95
class 3=215
class 4=109
class 5=2170
class 6=4015
As you can see that the data is highly unbalance. In testing mode I am almost missing class 2,3 and 4. How I can balance my classifier. Any hint will be highely appreciated.
Your data are multi-class (6- class) and you faced multi-class imbalance problem ( IR=79.59) follow these papers
and What does it mean: "How I can balance my classifier" because you need to balance your data set not your classifier!!!
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=5597016&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D5597016
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=4370991&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D4370991
https://www.cs.bham.ac.uk/~xin/papers/Shuo_SMCb-V3.pdf
I guess your ultimate goal is to reduce the misclassification rate (error rate) and all errors are equally important. In this case you are fine with the default behavior of packages like scikit learn. My suggestion is to play with the parameters. Make sure to also try Boosted Decision Trees. For a quick try you can use scikit learn. In my experience xgboost is the best boosted tree package with a nice python interface. (https://github.com/dmlc/xgboost)
You can refer following link for imbalance data classification.
http://sci2s.ugr.es/imbalanced
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training