I am trying to perform a multilevel stress classification. The existing literature reports the use of LDA, QDA, SVM-linear, SVM-rbf, kNN and NB for classification. But, are these classifiers competitive enough in accuracy when compared to algorithms such as gradient-boosted decision trees, random forest, or deep neural networks?
Anusha,
Loosely speaking, there is no such criterion. The reason is what you have already found: there are plenty of algorithms rather than one. To be clear, a classifier performance usually depends on the application and nature of your data. However, DNN almost all times performs competitive results, depends on how you represent the data to that.
Yon can use a lot of different complex criteria, based on Recall, Precision, TPR, FPR, etc. You also can use some own, special score and penalty oriented criteria - e.g. non-linear and non-symmetrical. And regular situation is following - according some criteria the SVM will be best, according another criteria the NN will be best,...
Classifier performance also depends on the preprocessing techniques like PCA,MA,DWT before being feed the data to algorithms. Data cleaning , visualization and pattern analysis also matters a lot . Test and Train set distribution also matters.
- Badges
- Science topic
- Similar topics
- Biological Science
- Physiology
- Physiological Processes
- Stress