H. Deng, "Interpreting Tree Ensembles with inTrees",
>> Available in inTrees R package. More informationH. Deng, "Guided Random Forest in the RRF Package"
>> More informationH. Deng, G. Runger, E. Tuv, W. Bannister, "CBC: An Associative Classifier with A Small Number of Rules >> This paper provides some insights/examples about decision trees and associative classifiers. >> "I had the opportunity to read a worthy research work addressing the issue of associative classifiers." Comments from DSSH. Deng, G. Runger, "Gene Selection with Guided Regularized Random Fores >> The "RRF" R package implements the (guided) regularized random forest algorithm, which is a type of regularized trees. >> More about RRF (illustrative examples, clarifications, etc.) H. Deng, M. Baydogan, G. Runger, "SMT: Sparse Multivariate Tree", >> Matlab+C code. (thanks to Mustafa for preparing this version for public) >> SMT finds a linear combination of a small subset of variables at each node. In addition to regular data, it can also be used for time series data and can generate informative temporal patterns.H. Deng, G. Runger, E. Tuv, M. Vladimir, "A Time Series Forest for Classification and Feature Extraction >> TSF provides the temporal importance curve for understanding the temporal patterns useful for classification. >> TSF outperforms NN with Euclidean distance or DTW. >> The time series must be of the same length. Otherwise need to align the time series into the same length. >> note: the error rate of SonyAIBORobotSurface and SonyAIBORobotSurfaceII should be switched in the paper. (thanks to Professor Eamonn for pointing it out)H. Deng, G. Runger, E. Tuv, >> R_Code>> "This is a really interesting approach with potential for wide application." Comments from JQT, a prestigious journal in the process monitoring area.H. Deng, G. Runger, "Feature Selection via Regularized Trees", Proceedings of the 2012 International Joint Conference on Neural Networks (IJCNN), IEEE, 2012.
>>The relationship between regularized trees and ordinary trees is similar to the relationship between Lasso and ordinary regression. Note Lasso is a linear model, while regularized trees can capture non-linear interactions between variables, and naturally handle missing values, different scales, and numerical and categorical variables.H. Deng, G. Runger, E. Tuv, "Bias of Importance Measures for Multi-Valued Attributes and Solutions >> R_Code >> Quite a few variable importance measures e.g. tree-based models are biased for multi-valued attributes. "Partial permutation" and "OOB Forest" were used to reduce the bias. H. Deng, S. Davila, G. Runger, E. Tuv,
Others S. Ji, A. Fakhry, and H. D., S. H., W. J., and H. D., "SVM in RTC", E. A., and H. D., "Predict disease in Crop", |