I'm designing a prediction model for the classification of milling cutter conditions based on vibrations evolved during the tool-workpiece interface using ML.
For this case, Random forest and SVM always exhibited the highest accuracy of classification over other classifiers (without even tuning of hyper parameters). Why is that? What is the specality?
Dear Abhishek D. Patange ,
as for support vector machine (SVM) is a powerful algorithm since the objective of such a model is not only to separate the classes but also to maximize the margin between them, thus maximizing their separation. This results in a high generalization capability.
As for random forest (RF), this algorithm takes advantage of a simple model as a decision tree (DT), while overcoming its poor generalization capability. Specifically, DT creates a very shallow margin between classes, thus often tends to overfit. RF trains several DT at the same time, feeding them with different subsets of features (so, only a subset of the original dataset is input to each decision tree). Then, the final output is computed using majority voting (each DT model outputs a prediction, then RF outputs the most frequent prediction among all DT models). This prevents overfitting and makes RF one of the most powerful models between shallow machine learning algorithms.
"No free lunch theorem". May be your data fits in that way. Have you tried various data divison combinations? Or k-fold validation?
@Nouman Amjad Raja, Yes I have tried different combinations and K fold cross validation also.
Have you tried to use automated system-cognitive analysis and its software tools - the intelligent Eidos system-to solve this problem? I am the developer of this artificial intelligence technology and I am ready to help you use it to solve your problem.
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification