I have developed an image classification model that categorizes images into two classes (we'll say good and bad for the sake of example) based on a set of tags. To improve the model's performance, I performed feature selection, which resulted in a subset of informative features. However, I encountered a challenge when dealing with a new batch of images.

The issue is that some features used in the initial feature selection process are present only in the first dataset, while other features are exclusive to the new batch. To address this discrepancy, I decided to select the features that are common to both sets (the intersection of the two feature sets) and retrain the model using these shared features.

While this approach seems logical to me, I would appreciate additional perspectives and insights on how to handle this situation more effectively. Is selecting the intersection of features a valid approach, or are there alternative methods or considerations that I should be aware of? I am open to suggestions and recommendations to improve the feature selection process and ensure robust classification performance for the new batch of images.

Thank you in advance for your insights and advice!



You must log in to answer this question.