I am using the e1071 package to train an SVM model with the argument "probability=TRUE". I then use "predict" with "probabilities = TRUE" and get the probabilities for the data point belonging to either class. So far all is well.
I think it's due to the stochastic nature of the algorithm. This guide might help you to understand it https://machinelearningmastery.com/stochastic-in-machine-learning/
As I said before, everything is exactly the same. The training set is the same too. However, the results are differents. You can test it in R software. Im confused.
I looked at these explanations. SVM uses a cross-validation step in developing the estimates of probabilities. So the probabilities in SVM dosen't have “transportability” and "reproducibility". We can not use the best cut-off point of a variable in other data with the similar variable.