It's a proposed classifier algorithm that is suppose to be better than Random Forest and XGBoost for classifying high dimensional data. The data sets are cancer data (prostate and myeloma). Unfortunately it's not going to be publish because I'd like to graduate sooner and that the software does not meet certain criteria for the journal we were aiming for.
The proposed algorithm uses two technique:
1. My forest consist of GUIDE decision trees by Dr. Loh. It is better than CART and M4.5 and such because it does not have the selection bias problem. CART and M4.5 are bias to selecting categorical predictor for node splitting. They're also bias on variables that enable more splitting so decision trees usually contain more levels. GUIDE is also aim at finding interactions candidate to split if it is statistically significant.
2. CERP by Dr. Moon. It makes the trees within the forest less correlated among each other. Much more so than Random Forest. Accuracy takes a hit as your correlation gets higher (obviously zero is the best). It also enabled ensemble of ensembles (ensemble of forests). Of course some of you may state that you can do ensemble of random forests but it is naive and won't help you.
I should be defending in this month or next month.