1

Is it possible to customize the rules in Random Forest algorithm for tree species classification?

all the integration or customization of rules in random forest algorithm is done in medical field, crime, drugs.

I have never found it done for tree species.

If it is possible can anyone suggest the suitable software

Improve this question
4
  • 1
    Can you please cite your sources? RF estimates are based on plurality of votes across trees. The entire premise of RF is weak-learning and convergence using Bootstraping. If you "customize" splitting rules you would severely bias the model and defeat the point of the model. The class-level probabilities would no longer converge on the sample. Once you start weighting trees you effectively decompose into a boosting model and not RF. I would recommend looking at model specification and not modifying the underlying algorithm.
    – Jeffrey Evans
    Commented Feb 2, 2015 at 16:20
  • Evans, J.S. and S.A. Cushman (2009) Gradient Modeling of Conifer Species Using Random Forest. Landscape Ecology 5:673-683. - and - Evans J.S., M.A. Murphy, Z.A. Holden, S.A. Cushman (2011). Modeling species distribution and change using Random Forests in Predictive species and habitat modeling in landscape ecology: concepts and applications. eds Drew CA, YF Wiersma, F Huettmann. Springer, NY
    – Jeffrey Evans
    Commented Feb 2, 2015 at 16:23
  • One of the paper that i read integrated random forest rules and use it to replace the original rules. It is done in medical field
    – Atikah
    Commented Feb 11, 2015 at 2:14
  • Sirikulviriya,N and Sinthupiny,S (2011) Integration of rules from a Random Forest. 2011 International Conference on Information and Electronic Engineering
    – Atikah
    Commented Feb 11, 2015 at 2:16

1 Answer 1

Reset to default
2

The RandomForests algorithm is often used in forestry. There are two implementations of the randomForests algorithm that I regularly use. The first is a pixel-based classifier implimented in R using the randomForests package. I believe this is most sophisticated and flexible approach you are likely to find. There are a many resources to get you started using this route. The following publication used a variety of tree detection algorithms, including randomForests, to identify juniper cover:

Poznanovic, A. J., Falkowski, M. J., Maclean, A. L., Smith, A. M., & Evans, J. S. (2014). An Accuracy Assessment of Tree Detection Algorithms in Juniper Woodlands. Photogrammetric Engineering & Remote Sensing, 80(7), 627-637.

Here are some additional publications that may be of interest:

Immitzer, M., Atzberger, C., & Koukal, T. (2012). Tree species classification with random forest using very high spatial resolution 8-band WorldView-2 satellite data. Remote Sensing, 4(9), 2661-2693.

Rodriguez-Galiano, V. F., Ghimire, B., Rogan, J., Chica-Olmo, M., & Rigol-Sanchez, J. P. (2012). An assessment of the effectiveness of a random forest classifier for land-cover classification. ISPRS Journal of Photogrammetry and Remote Sensing, 67, 93-104.

The second approach is a object-oriented implementation of the randomForests algorithm in eCognition Developer (There is actually a pixel-based implimentation too). This is a very powerful approach, although is very difficult to implement as there is little documentation. the best advice I can give you is to join the eCognition community and search for "CART, SVM & RF Classifier Example (Improved in eCognition 9.0)" to find the example ruleset. This example with sample imagery included will walk you through how to perform a pixel-based or object-oriented randomForest classifcation.

Improve this answer
2
  • I would limit application of SVM's to binomial problems.
    – Jeffrey Evans
    Commented Feb 2, 2015 at 16:21
  • 1
    @Aaron: Thank you. I was interested to use rule based method in classifying tree species but I don't want to use/create model. That is why I want to try to "change" rules in random forest. Do you have any suggestions on how to incorporate rules based method and random forest? Thank you
    – Atikah
    Commented Feb 11, 2015 at 2:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.