Hi Joshua,
SVMs use kernels and these kernel methods are based on distance. So, an SVM assumes that the data it works with is in a standard range, usually either 0 to 1, or -1 to 1. As a result, if the features are not scaled to comparable ranges, the features with the largest range will completely dominate in the computation of the kernel matrix. This appears to be the reason as to why you are not getting expected predicted values since the two features ‘alpha’ and ‘beta’ have different scales as can be seen in the below plot:
>> plot(X, 'LineWidth', 3)
One way to improve the results is to normalize/standardize the features in data ‘X’ prior to using this data for training an SVM model. To do so, you should use the ‘Standardize’ parameter flag with value TRUE as a Name-Value pair argument passed to the function ‘fitrsvm’ as follows:
>> PasMdl = fitrsvm(X,SfPas', 'Standardize', true);
For more information about the use of ‘Standardize’ flag, please check the following link: https://www.mathworks.com/help/stats/fitrsvm.html#busljl4-Standardize
An example where ‘Standardize’ flag is used can be checked here: https://www.mathworks.com/help/stats/fitrsvm.html#buxgtub
Alternatively, you can yourselves normalize/standardize the data as part of data pre-processing using the data normalization function ‘normalize’: https://www.mathworks.com/help/matlab/ref/normalize.html
Hope this helps.
Thanks.
