I have a question about the correct way to use the SMOTE sampling algorithm. I have been read a lot about this algorithm. I forced to use SMOTE within my code, so I can't use any tools like KNIME or WEKA. After few days in searching, I can say that there are two implementation of SMOTE, one in R language and other included in weka java library. I have downloaded the weka.dev.jar file and weka.core.jar file and used them in my code. Unfortunately I'm not expert enough in Java. Can anyone help me. I just need to use SMOTE with crossvalidation. and both SMOTE and crossvalidation are java clasess. But I failed to write a simple java class that create objects from these classes. The reaseon is that theses classes need another objects from other classes.
any help will be appreciated
Dear Ahmed,
Please go to the following links:
https://www.codeproject.com/Questions/1231814/How-to-use-weka-smote-smapling-algorithm-in-java-c
https://github.com/bnjmn/weka/blob/master/packages/external/SMOTE/src/main/java/weka/filters/supervised/instance/SMOTE.java
https://www3.nd.edu/~dial/software/weka/smote.java
Thanks,
Sobhan
Dear Ahmed,
Please go to the following links:
https://www.codeproject.com/Questions/1231814/How-to-use-weka-smote-smapling-algorithm-in-java-c
https://github.com/bnjmn/weka/blob/master/packages/external/SMOTE/src/main/java/weka/filters/supervised/instance/SMOTE.java
https://www3.nd.edu/~dial/software/weka/smote.java
Thanks,
Sobhan
Sobhan,
I appreciate your reply, but let me correct it.
I'm a member of WEKA environment and I can say that non of your links are applying the SMOTE in conjunction with the classification algorithm as well as cross-validation properly.
The best way to solve the problem of Hussein is to run SMOTE within FilteredClassifier (the meta-learner algorithm).
HTH.
Samer Sarsam, PhD.
Try with the following link:
https://www.codeproject.com/Questions/1231814/How-to-use-weka-smote-smapling-algorithm-in-java-c
https://github.com/bnjmn/weka/blob/master/packages/external/SMOTE/src/main/java/weka/filters/supervised/instance/SMOTE.java
Mr. Samer,
I have a question, when applying SMOTE in the experimenter or the knowledge flow, is the oversampling of the minority class implemented on the entire training and testing datasets, or is it done on the training dataset only ? As far as i know, SMOTE should only be applied on the training set, while maintaining the testing set as is. It would be greatly appreciated if you elaborate on that point.
Mohammad Ahmad,
Always, the SMOTE filter has to be applied *only* on the training set, but the problem occurs when this training set itself is getting splitted later into training and test sets using, e.g., cross-validation/holdout method. This will bring some information of the class into the training set (due to the nature of the SMOTE filter) as well as the test set (part of the original training set), which produces an optimistic result (this is cheating!). Thus, in my previous post, I recommended using the SMOTE filter through “FilteredClassifier” to keep the test set safe without changing its structure. The recommended approach can be applied at any of WEKA’s platforms (Explorer, Experimenter, and KnowledgeFlow). FilteredClassifier allows running an arbitrary classifier on data that have been passed through an arbitrary filter. Like the classifier, the structure of the filter is based exclusively on the training data and test instances will be processed by the filter without changing their structure.
In the Experimenter, the “Simple” mode, FilteredClassifier (and the SMOTE filter though) can applied under the “Algorithms” panel, while in the “Advanced” mode, SMOTE can be applied through FilteredClassifier under “ClassifierSplitEvaluator” (a parameter of CrossValidationResultProducer).
In KnowledgeFlow, as highlighted up, you can utilize the SMOTE filter as a parameter of “FilteredClassifier”, so you need to use the “FilteredClassifier” component; otherwise, the SMOTE filter component has to be placed after the “CrossValidationFoldMaker” component, if you managed to use cross-validation, for instance. If the SMOTE is going to be used before the CrossValidationFoldMaker that will bring some information of the class into the test set, which produces an optimistic result later on when classifying the data--the training set will carry some of the class information as well as the test set.
Dr. Samer Sarsam
Dr. Samer,
Thank you very much for your input and explanation. I will get a chance to repeat my experiments again something over the next week or two.
If you have weka installed in your pc then simply go to tool and add library SMOTE. then go to your installation folder where you insatll weka and find the weka folder there.
Or simply dwnld the smote jar file from github, java2s or from simply search google, you must found the jar file.
Got the library from here
https://mvnrepository.com/artifact/nz.ac.waikato.cms.weka/SMOTE/1.0.2
Currently, not included in the standard WEKA 3.8
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining