How to use Parallel Coordinates Plot for Predictor selection?

Hi,

I have a question about Parallel Coordinates Plot from Classifier app (Machine Learning).

I have Parallel Coordinates Plot just like the one shown in the figure on this page: https://www.mathworks.com/help/stats/feature-selection-and-feature-transformation.html#buwh6hc-1

In the description of this page (on point 5), its mentioned that "If you identify predictors that are not useful for separating out classes, use Feature Selection to remove them and train classifiers including only the most useful predictors."

Its not clear to me how can I use this plot to figure out which predictior are not useful for separating out the classes? In my plot I have 35 features for 2 classes, I want to remove the features which are not helpful for disntnigushing my classes, so I want to reduce the dimensionality of my data and remove the unuseful features. But I have to idea how this figure can be helpful me in removing those features.

Any help would be really appreciated.

Thanks !

Sahil

Answer 1

[Patel Mounika]

Let’s look at the parallel coordinate plot shown in the figure you sent: https://www.mathworks.com/help/stats/feature-selection-and-feature-transformation.html#buwh6hc-1

In this plot take look at the comparison of the sepal widths of the different flowers, the values for setosa, versicolor and virginica are overlapping because of which it will be difficult to classify based on this feature and on the other hand the values of petal width are distinct for different flowers (or not overlapping) which will help in classification of the flowers. So, based on this petal length and petal width are the features that separate the classes best compared to sepal width and sepal length.

Hope this helps.

[Sahil Bajaj]

Thank you so much Mounika, that really helps.

Answer 2

[Perry Gogas]

But I think that you also have to look at the missclassified cases marked with the dashed lines. These too provide information on the importance of each variable.