How do you handle imbalanced datasets when training machine learning models, and what techniques have proven most effective in improving model performance?

When I deal with imbalanced datasets, my go-to method is class weighting. I find it straightforward and really effective. I sometimes consider oversampling too, but I'm not a big fan because it can make the data noisy. Undersampling? I tend to avoid it because it feels like you're just throwing away too much useful information. Overall, class weighting just seems to strike the right balance and helps my models perform better without messing up the data.