Classification of overlapping classed

Hello everyone.

Iam quite new to Machine learning and played around with some of the standard datasets. Now I wanted to use my own one, which includes two features only and two classes (0,1).

The problem is that the data distribution looks like an exponential function and the two features overlapp.

All example datasets I used so far have been linearly seperable.

Now Iam wondering which type of machine learning algorithm would be best in auch cases?

Any advice would be really appriciated.

Thanks in advance,

Kind Regards,

Mike

Answer 1

[Von Duesenberg]

If you have the Statistics and Machine Learning Toolbox, the easiest way to get you started is to run several classifiers with the Classification Learner App. QDA or SVM might be good options but it's hard to tell because it very much depends on your data.

As a side note, it may be that some of your examples "belong" to both class 0 and class 1 to some extent, in which case you may want to reflect this degree of membership with things like fuzzy c-means clustering.

If your features really overlap too much, a possible solution would be to work with a "raw" representation of the data (e.g. an image, no explicit feature) and let a neural network (like a CNN) extract relevant features for you.