Deep forest model development and training

1. A computer implemented method for developing and training models for analyzing data, the method comprising:

constructing a model by:

growing, by one or more computer processors, a random forest of decision trees from a data set;

determining, by the one or more computer processors, Out-of-bag (OOB) predictions for an instance of the data set for the random forest;

appending the Out-of-bag predictions as a probability vector to the data set as new features represented in columns for the instance of the data set by the one or more computer processors;

growing, by the one or more computer processors, an additional random forest using the data set with the appended OOB predictions; and

combining an output of the additional random forest, by the one or more computer processors, with a combiner.

2. The computer implemented method according to claim 1 wherein each random forest is grown from data selected from a group consisting of: numeric, text, audio, video, image data location, speech, music, entertainment, healthcare, financial information, vehicle, logistics, and sales data.

3. The computer implemented method according to claim 1 , wherein growing the random forest of decision trees using the data set comprises growing decision trees using a bootstrapped sample, taken with replacement from the data set, to grow each tree.

4. The computer implemented method according to claim 1 further comprising determining an OOB accuracy for each random forest layer and adding random forests until the OOB accuracy improves less than 0.005% from an OOB accuracy of a preceding layer.

5. The computer implemented method according to claim 1 wherein the combiner comprises a structure selected from a group consisting of: a random forest, and a gradient boosting structure.

6. The computer implemented method according to claim 1 wherein the model comprises a sequence of layers, each layer comprising a single random forest.

7. A computer program product for developing and training models for analyzing data, the computer program product comprising one or more computer readable storage media and program instructions stored on the one or more computer readable storage media, the program instructions comprising:

programmed instructions for constructing a model by:

providing a data set;

growing a random forest of decision trees from the data set;

determining Out-of-bag (OOB) predictions for an instance of the data set for the random forest;

appending the Out-of-bag predictions as a probability vector to the data set as new features represented in columns for the instance of the data set;

growing an additional random forest using the data set with the appended OOB predictions; and

combining an output of the additional random forest with a combiner.

8. The computer program product according to claim 7 , wherein providing the data set comprises providing a data set from a group consisting of: numeric, text, audio, video, image data location, speech, music, entertainment, healthcare, financial information, vehicle, logistics, and sales data.

9. The computer program product according to claim 7 , wherein growing the random forest of decision trees comprises growing decision trees using a bootstrapped sample, taken with replacement, from the data set to grow each tree.

10. The computer program product according to claim 7 , further comprising determining an OOB accuracy for each random forest layer and adding random forests until the OOB accuracy improves less than 0.005% from an OOB accuracy of a preceding layer.

11. The computer program product according to claim 7 , wherein the combiner comprises a structure selected from a group consisting of: a random forest, and a gradient boosting structure.

12. The computer program product according to claim 7 , wherein the model comprises a sequence of layers, each layer comprising a single random forest.

13. A computer system for developing and training models for analyzing data, the computer system comprising:

one or more computer processors;

one or more computer readable storage devices;

program instructions stored on the one or more computer readable storage devices for execution by the one or more computer processors, the program instructions comprising:

programmed instructions for constructing a model by:

providing a data set;

growing a random forest of decision trees from the data set;

determining an Out-of-bag (OOB) predictions for an instance of the data set for the random forest;

appending the Out-of-bag predictions as a probability vector to the data set as new features represented in columns for the instance of the data set;

growing an additional random forest using the data set with the appended OOB predictions; and

combining an output of the additional random forest with a combiner.

14. The computer system according to claim 13 , wherein providing the data set comprises providing a data set from a group consisting of: numeric, text, audio, video, image data location, speech, music, entertainment, healthcare, financial information, vehicle, logistics, and sales data.

15. The computer system according to claim 13 , wherein growing the random forest of decision trees comprises growing decision trees using a bootstrapped sample, taken with replacement, from the data set to grow each tree.

16. The computer system according to claim 13 , further comprising determining an OOB accuracy for each random forest layer and adding random forests until the OOB accuracy improves less than 0.005% from an OOB accuracy of a preceding layer.

17. The computer system according to claim 13 , wherein the combiner comprises a structure selected from a group consisting of: a random forest, and a gradient boosting structure.

18. The computer system according to claim 13 , wherein the model comprises a sequence of layers, each layer comprising a single random forest.

19. A computer implemented method for developing and training models for analyzing data, the method comprising:

constructing a model by:

receiving, by one or more computer processors, a training data set;

growing, by the one or more computer processors, a random forest of decision trees from the training data set;

determining, by the one or more computer processors, Out-of-bag (OOB) predictions for an instance of the data set for the random forest;

appending the Out-of-bag predictions as a probability vector to the training data set as new features represented in columns for the instance of the training data set by the one or more computer processors;

determining, by the one or more computer processors, a first OOB accuracy for the random forest;

growing, by the one or more computer processors, an additional random forest using the training data set with the appended OOB predictions;

determining, by the one or more computer processors, a second OOB accuracy for the additional random forest;

comparing, by the one or more computer processors, the first OOB accuracy of the random forest and the second OOB accuracy of the additional random forest; and

combining an output of the additional random forest by the one or more computer processors with a combiner.

20. The computer implemented method according to claim 19 wherein each random forest is grown from data selected from a group consisting of: numeric, text, audio, video, image data location, speech, music, entertainment, healthcare, financial information, vehicle, logistics, and sales.

21. The computer implemented method according to claim 19 , wherein growing the random forest of decision trees using the training data set comprises growing decision trees using a bootstrapped sample, taken with replacement from the training data set, to grow each tree.

22. The computer implemented method according to claim 19 , wherein the model comprises a sequence of layers, each layer comprising a single random forest.

23. A computer implemented method for developing and training models for analyzing data, the method comprising:

constructing a model of sequential layers, each layer including a single random forest, by:

receiving, by one or more computer processors, a training data set;

receiving, by the one or more computer processors, a determined number of trees per forest and a class vector specification;

growing, by the one or more computer processors, the number of determined trees for a first forest using the training data set;

determining, by the one or more computer processors, first Out-of-bag (OOB) predictions for an instance of the data set for the first forest;

appending the first OOB predictions as a probability vector to the training data set as new features represented in columns for the instance of the data set by the one or more computer processors;

determining an OOB accuracy for the first forest by the one or more computer processors;

growing the number of determined trees for an additional forest using the training data set with the appended first OOB predictions by the one or more computer processors;

determining an additional OOB prediction for the additional forest by the one or more computer processors;

appending additional OOB predictions to the training data set by the one or more computer processors;

determining an additional OOB accuracy for the additional forest by the one or more computer processors;

adding forests, by the one or more computer processors, until the additional OOB accuracy does not improve; and

combining an output of the additional forest by the one or more computer processors.

24. The computer implemented method according to claim 23 wherein each random forest is grown from data selected from a group consisting of: numeric, text, audio, video, image data location, speech, music, entertainment, healthcare, financial information, vehicle, logistics, and sales data.

25. The computer implemented method according to claim 23 , wherein growing the random forest of decision trees using the training data set comprises growing decision trees using a bootstrapped sample, taken with replacement from the training data set, to grow each tree.