Consider a scenario where you are developing a model to detect fraudulent transactions in a financial system. Given the concepts of confusion matrices, accuracy, precision, recall, F$-$measure, kappa statistic, sensitivity$,$ and specificity:

How would you evaluate the performance of your model using a confusion matrix?

Which performance metrics would be most important in this context, and why?

Discuss how the class imbalance in fraud detection $($where fraudulent transactions are much less frequent than non$-$fraudulent ones$)$ affects the evaluation of your model. What strategies could you use to address this imbalance in your performance assessment?

Explain how you would use cross$-$validation and bootstrap sampling to ensure your model's performance is reliable and not just a result of overfitting to your training data.