Do we need to normalize data before using matlab machine learning classifier especially SVM?

Hi,

i al currently exploring machine learning using matlab. My question is do i need to normalize the data between 0-1 before start training with machine learning classifier expecially SVM?

For example the image below, we notice that the data are between -5.2 to 6.5 for "X" and -6.4 to 5.6 for "Y". so do i need to normalize the data from 0 to 1 first before start to train with machine learning classifier such as SVM?

Besides, i do notice that SVM does have kernal function properties such as gaussian. Does it mean that using kernal function alone is sufficient to ovveride the normalization process?

Hope someone could help clarify the confusion. many thanks.

Answer 1

[Manas]

Hey Teo,

I understand that you wish to know if there is a need to normalize data when using MATLAB SVM Classifier

It is generally recommended to normalize the input data before training as it helps ensure that all features have a similar scale, which can improve the performance and convergence of the model.

While kernel functions like gaussian or linear can help handle non-linear data and implicitly map it to a higher-dimensional feature space, normalization still plays a crucial role in ensuring all features have a similar scale.

With all being said, the decision to normalize the data ultimately depends on the specific characteristics of the dataset and the requirements of the problem statement itself. I would recommend experimenting with both normalized and unnormalized data to observe the impact on the performance of your SVM Model.

I’m adding links to relevant documentations for your reference:

You can learn more about SVM’s from here: https://in.mathworks.com/discovery/support-vector-machine.html

SVM Documentation for MATLAB: https://in.mathworks.com/help/stats/support-vector-machine-classification.html

Normalization: https://in.mathworks.com/help/matlab/ref/double.normalize.html

Hope this helps