I have applied genetic algorithm with SVM as feature selection method for high dimensional datasets (i.e gene expression data). The algorithm was evaluated based on classification accuracy using 10 folds cross validation.
There exist state-of-the-art methods which are applied on the same datasets. For Fair comparison, can i reported there results and compare with them directly on my research. Note that all of them used same validation (i.e 10 folds cross validation), its mean that It means the input data has different set on each experiment. Is this mean that comparison will not be fair ???
Should I have re-implemented the methods in the literature to do a fair comparison.
Thanks in advance
Hi,
As far as I know, If the data set and the evaluation strategy are not changed, it is not necessary to re-implement the other methods. Just report your results.
Hi Osama,
You do not have to re-implement the methods in the literature. If your experiments using exactly the same dataset and use the same evaluation technique which is 10-fold cross validation, these two are enough to compare your method with the state-of-the-art.
' It means the input data has different set on each experiment' > this doesn't matter because you are doing 10-fold cross validation means all training and test data for each fold are randomly selected.
however, you should be careful about the parameters the méthodes in the literature used. for example, for fair comparaison you shoud use the same population size and the same number of evaluations for all the algorithms. Also, there are some works who use the same numbre of iterations instead of number of evaluation.
some works use same initial solution in the case of population based methods like Genetic algorithm.
- Badges
- Science topic
- Similar topics
- Biological Science
- Genetics
- Gene Expression