I was performing the both classifiers for two different hyperspectral datasets. For one dataset SVM is performing better and for the other one RF is performing better. What could be the probable reasons for this results!
Dear Subir,
It depends on the data, we can't claim whether SVM is better or RF.
Try to decide by analyzing box plots on the two data sets. If there is a clear discrimination between two datasets, then SVM will work better. If the dataset is overlapping to some extent, then RF may give better results.
Thanks
Dear Subir
I am not working on your field. But I found some of these papers which might help you. The links of the papers are as follows:
1. http://www.sciencedirect.com/science/article/pii/S0924271611001304
2. http://www.sciencedirect.com/science/article/pii/S0924271616000265
3. http://www.tandfonline.com/doi/abs/10.1080/01431160412331269698
4. http://www.sciencedirect.com/science/article/pii/S0924271610001140
5. http://ieeexplore.ieee.org/document/1323134/#full-text-section
6. http://www.mitpressjournals.org/doi/abs/10.1162/089976698300017197
7. http://www.tandfonline.com/doi/abs/10.1080/01431161.2014.903435
8. http://www.mdpi.com/1424-8220/12/4/4764/htm
9. http://www.tandfonline.com/doi/abs/10.1080/01431160903140803
10. https://www.researchgate.net/profile/Ingmar_Nitze/publication/275641579_COMPARISON_OF_MACHINE_LEARNING_ALGORITHMS_RANDOM_FOREST_ARTIFICIAL_NEURAL_NETWORK_AND_SUPPORT_VECTOR_MACHINE_TO_MAXIMUM_LIKELIHOOD_FOR_SUPERVISED_CROP_TYPE_CLASSIFICATION/links/5541238e0cf2b790436bc791.pdf
Regards
Pankaj
Conference Paper COMPARISON OF MACHINE LEARNING ALGORITHMS RANDOM FOREST, ART...
I confirm it depends on the data set.
You can tune your classifiers using cross validation and then chose the best classifier and the best parameters.
Dear Subir,
First, study the basics of machine learning. Each method has different characteristics. Well, it is really up to the learning process of each method. However SVM is more popular for the analysis of hyperspectral data. The underlying reason of SVM’s popularity for classification is achieving the high classification accuracy with a small number of training data and able to outperform than other conventional methods such as ANN and ML classification (not everytime, of course). However in some cases, RF can outperform SVM depending upon the input data and training data.
Therefore we can not say any of them is more powerful. Also look at the "no free lunch theorem" then you can understand better what we say. Hope it is clear.
See this paper: Ustuner, M., Sanli, F. B., and Abdikan, S.: BALANCED VS IMBALANCED TRAINING DATA: CLASSIFYING RAPIDEYE DATA WITH SUPPORT VECTOR MACHINES, Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B7, 379-384, https://doi.org/10.5194/isprs-archives-XLI-B7-379-2016, 2016.
https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLI-B7/379/2016/isprs-archives-XLI-B7-379-2016.pdf
Thank you Sir (Mustafa Üstüner) for your Response. I will consider your suggestions though some parts are already known to me!
It actually depends on the data set used, but mostly Random Forests acts better for classification, whereas Support Vector Machine acts better in regression. Try to tune the variables of RF seeking better results.
Hi Subir,
I can also confirm that the performance of each method highly depends on the dataset. Make sure that you tune the parameters of your classifiers. In my experience SVM is a bit more tricky in that regard and needs a bit more pre-processing (choosing the right kernel, etc). This might also be a reason for the difference in performance.
In our paper, which Pankaj Dey referenced (#10) SVM produced the best results, but I did not use hyperspectral data. RF was not as good, but easier to handle. In other projects RF has performed better.
Thank You Sir (Ingmar Nitze) for your response. I read few papers and most of them are concluding the same that it will depend on the dataset.
Performance of the classifier depends up on the dataset and cross-validation. In our work SVM gives the best results compared to other classifiers.
It actually depends upon the type (Text/Images) and the size of the dataset. But, in most cases, SVM under different parameter settings can give better results.
Depends upon the database and also feature Extraction. So try your algorithm on different learning techniques and choose best classifier.
Article Support Vector Machine vs. Random Forest for Remote Sensing ...
- Badges
- Science topic
- Similar topics
- Clinical Pharmacology
- Dosing