Classification of a noisy data

Question

What method can be used to classify data in the following example? There is a table (hundreds of strings and hundreds of columns). Several columns in this table uniquely allow you to classify each row:

Class 0 : …noisy bits…00…noisy bits…

Class 0 : …noisy bits…11…noisy bits…

Class 1 : …noisy bits…10…noisy bits…

Class 1 : …noisy bits…01…noisy bits…

If bits in columns are 00 and 11, this is in class 0. If bits in columns are 10 and 01, this is in class 1.

The problem is we don’t know the numbers of these columns, their number (could be 2 or 10), and the combination of bits.

Is there a method that can learn to classify this?

I tried SVC with different parameters, simple NN, and even Apriori. Works nothing.

Here are my experiments:

Prepare data files for experiments: https://github.com/sdiving777/SVC/blob/main/svc_test_v2_prepare_files.py

Simple classification, that shows that it’s possible to classify: https://github.com/sdiving777/SVC/blob/main/svc_test_v2_simple_check.py

SVC test: https://github.com/sdiving777/SVC/blob/main/svc_test_v2_train_and_check.py

NN test: https://github.com/sdiving777/SVC/blob/main/NN_test_v2.py

Apriori test: https://github.com/sdiving777/SVC/blob/main/apriory_test.py

Answer 1

You should apply some sort of searching/cleaning first...try finding any sort of garbage you can rule out.

The question here is to find 00,11,01,10 among noisy bits (which too are binary, hence heavily confusing). Any Classification algorithm (like SVM) will need a more clear distinction.