Hey KNIME Community,
i have a question regarding to the results of my algorithms or better said regarding to the comparison of the algorithms which you can see in the following picture
In the literature i have read that in general the SVM is the best algorithm for classification tasks, but my results show me something different.
What is the reason for the different results?
Is it true that you can not always use all the algorithms for a particular problem?
Thanks and kind regards,
hm I guess that statement was made by someone who developed SVM based approaches
If you take a look at Kaggle (an online platform for data mining challenges) you will find that there are two rough categories of problems that are solved in either of two ways:
- Problems with structured (for lack of a better word) data, e.g. computer vision or text processing. These are usually tackled using deep learning
- Problems with tabular data e.g. customer records. These are often tackled using ensembles (and ensembles of ensembles) where the most popular algorithm is probably XGBoost (which is a special variant of Gradient Boosted Trees)
I guess your problem falls into the second category and as Gradient Boosted Trees have the second best accuracy this seems to be about right. The fact that the Tree Ensemble (aka Random Forest) beats the Gradient Boosted Trees is probably because they tend to be more robust with regard to their hyperparameter configuration. However, it is also always a good idea to test Random Forests as classifier because at the very least they will provide you with a good baseline without much (or any) hyper parameter tuning.
This is not to say that SVMs are bad algorithms but they are no longer as popular as they used to be, which is mostly due to the fact that they are more sensitive towards the choice of hyper parameters (e.g. which kernel to use) than e.g. Random Forests.
Please take what I wrote only as a rough guidance because in the end your problem will dictate which algorithm is best suited for it (and you will hopefully find that KNIME makes it easy to try out different approaches).
thank you very much for you detailed answer, that helps me a lot
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.