Classification problem with separate files

Hello everybody,

I am trying to solve a classification problem with multiple input files which are different examples for the problem. Every xlsx file contains different information about pores (size, shape, place) in a specimen made of different materials. I have about 100 files with minimum 4000 pores in each file. One of the pores is the critical pore which leads to failure when the specimen is loaded. That means that only one pore is classified as critical, the others are classified as uncritical. The task is now, to find the critical pore in the data with the help of machine learning. I cannot combine the seperate files since for each specimen the pore distribution is a little bit different.

I hope the problem got clear.

Thanks for your answer.

[Hiro Yoshino]

It seems that this is not a classification problem but an anomaly/outlier detection problem.

I bet the number of the critical pores is much smaller than that of other, i.e., the numbers are not well-balanced.

For this, there are some approaches. Some are based on probability distribution estimation and some are on boundary construction.

You should take a look at this site to see what would suit your case:

https://jp.mathworks.com/discovery/anomaly-detection.html

I would reccomend using one-class SVM as a starter since this is not based on probability distribution.

[Alexander Koch]

Thank you! That was really helpful.