Hi
I am working on classification task, my data contains numeric and categorical features (mixed), for classifying such data we need to use transformation method, for converting any type of data to another type, to be able to apply the exist classification algorithms such as SVM, Naive bayes or KNN .
My question is there any classification algorithm can handle mixed data without using any transformation method?
I think that without any form of data transformation you will not be able to solve the classification problem. Therefore, I suggest to consider the coding of the nominal data contained in the article: https://www.researchgate.net/publication/278403191_Numerical_Coding_of_Nominal_Data
In this way, you will be able to use the Euclidean metric to measure the distance between points in the data space.
In particular tree based methods do well with this sort of data because they branch on discriminative features. It's a little hard to say without knowing your specific application, but in general if you think that some of your features might be significantly more discriminative than others, you could try some dimensionality reduction techniques to clean this up a bit.
Also if you use a dimensionality reduction technique you end up getting a slightly more robust format for your feature vector (they generally end up being straight numerical vectors instead of mixed data types), which might let you leverage different methods. You could also look into hand engineering features. With properly hand engineered features Random Forest will get you very close to state of the art on most tasks.
I think there is no solution for ignoring the mentioned preprocessing step. Therefore, the symbolic-valued features (categorical features) should be mapped to numeric values or vectors. Notice, generally, dummy coding and label coding are two approaches that have been used in the reliable researches for this transformation.
Since the data type is mixed, you can use any nonlinear classifier such as Quadratic classifier, discriminant function analysis, tree-based classifier.
I think that if the features descriptors of your data are discriminative for each class, and if the number of your features descriptors are not too many, it is possible to not need for transformation method. but the choice of the type of classifier method depend on the separability of theses classes in their features (linear or nonlinear)
in my view may be i am not sure Random Forest may helpful, in this case.
- Badges
- Science topic
- Similar topics
- Philosophy
- Ethics