Regarding usage of classifier in case of imbalance data

I have images database consisting of four classes, but highly imbalanced. One class contains 150 images other 1500, 1800,1000. which classifier can be used to get efficient results for classification. How can i divide these images into training and testing. Kindly help

Answer 1

[Image Analyst]

You can use randsample() or randperm() to get a list of 120 (80% of the smallest group) from each of the 4 groups. Use those 480 images for training, and the rest for testing/validation of the model.

[varinder kaur]

Thanks for answer. But i want to perform 50: 50, 60: 40 division of training and testing data. In that case which method can i use to deal with imbalance data. also Can you please suggest classifier.

[Image Analyst]

You can decide how many samples you want for training and testing and use randsample. There's no way anyone can suggest a classifier for your data without seeing it. The best option is to use the Classification Learner App, on the Apps tab of the tool ribbon, and try them all and pick the one with the lowest error.

[varinder kaur]

Thanks Image Analyst for the answer

[Rocío García Mojón]

Hi, I have a similar problem to varinder kaur. I have a photo type database and these are unbalanced. It is about medical images. On one side, I have 655 pictures of the healthy type and 16000 of the diseased type. I would like to increase the number of pictures in the minority group, but so far I have not found a good way to do it.

Thanks in advance

Regards