Whenever you're ready, walk me through how you'd approach this.
How do you handle class imbalance when training a model on rare events, such as system failures or fraudulent transactions?
Explain how you would prepare the data, choose between class weighting and resampling, select evaluation metrics, validate the model, and choose a production decision threshold. Include the tradeoffs between false positives and false negatives, and describe how you would monitor performance after deployment.