Imagine you are solving a fraud detection problem where the target class $($a fraudulent activity$)$ is observed in $3\%$ of the training sample. Your model has $97\%$ accuracy after making predictions using the validation sample. Which metric is preferred in assessing the model's performance?