# Classification problem using features with unequal sizes

I am relatively new to Machine Learning/ Deep Learning and currently I am working on a classification problem. I have many 2D images and each of them is a cross section of a specimen showing the deformation after applying a load. The images show the edges of the specimen with the same color and the deformed edges are shown with different colors. I want to use the values of pixels as one of the feature to my classification problem. At the same time I also want to use the load value, the co-ordinates of the position of impact as features. Therefore my problem would look something like this

     Feature 1           Feature 2          Feature 3         Class
Vector of size N*1  Vector of size 3*1   Scalar value m1     Class A
Vector of size N*1  Vector of size 3*1   Scalar value m2     Class B


and so on..

I would like to know if there is a way to deal with this kind of problem or if there are any papers which focus on approaches to deal something similar to this. Any kind of help is really appreciated. If any clarification is needed from my side, do let me know. Thank you.

• The general approach would be to simply concatenate all the vectors into one long vector used as the instance. – Erwan Jul 16 '19 at 15:01

I am not sure if what I am gonna say is totally correct but you could try it. So instead of treating each of the pixels as part of one feature, you could append everything to make one array. That array would be of size N+3+1. So this would be the size of one row in data matrix and further you could provide weightage to the elements. Like if pixels are more important then mulitply them by some factor. One problem that I see in dealing with this might be standardization of data matrix.

• Thanks for the reply. What if I standardize the data first and then append everything. Does that make sense? – Guest_10215 Jul 24 '19 at 15:30
• Standardization would only scale things so I don't think it would affect the size you are looking at. – Parijat Bhatt Jul 24 '19 at 18:04