Nima Karimi if i'm in your place, i used either random forest or J48 for this problem.

thanks, Argha, random forest seems to be interesting, I can apply the decision tree model in Envi and now I see that random forest pick the most probable one among decision trees and that is one step forward.

is this any training resources on that?

embedded python in Arc-Gis and or R appears the most practical ones in such a machine learning method.

thanks

Dear Nima Karimi ,

Look the link, maybe useful.

https://www.analyticsvidhya.com/blog/2016/11/an-introduction-to-clustering-and-different-methods-of-clustering/

Regards,

Shafagat

thanks, Dr Shafagat Mahmudova

this really helps to understand the concept behind all those classification methods, Since I have specific features and my purpose is to target all of my interested features, I think I should use supervised classification after using a clustering method to enhance its accuracy like K-means.

again thanks for this article

METHOD OF COGNITIVE CLUSTERIZATION OR CLUSTERIZATION ON THE BASIS OF KNOWLEDGE (Clusterization in system-cognitive analysis and intellectual system "Eidos")

Article METHOD OF COGNITIVE CLUSTERIZATION OR CLUSTERIZATION ON THE ...

Nima Karimi for that you can utilize the option "supplied test set" in Weka.

The best source as a book to your query is here : https://cds.cern.ch/record/998831/files/9780387310732_TOC.pdf

I think if you combine K-means with a supervised algorithm you should be able to get good results. For the training of a supervised classification model you must split your data into training data and test data, a good rule of thumb is 70% and 30%, respectively (or 80%-20%). Let's say you have 1000 images you can split them to 700 images for training and 300 images for testing.

First, try using the data for one year and then if you already have data for another year, you can put them all together and split them, as I mentioned above. Using more data can sometimes improve your results.

Random Forest is generally one the best classification algorithms. Depending on the data that you another algorithm may outperform it.

So I'll advice you to test different algortihms that you prior have tuned and select the best of them

based on my experience, i suggest you the methods of SVM (support vector machine) and ANN (Artificial neural networks) .

K means may be a good option since it easily adapts to new samples.