Model Evaluation

Karina is evaluating three different supervised machine learning models for detecting fraudulent bank transactions. Karina knows that one of the weaknesses of the training data was that there are many more examples of valid transactions compared to fraudulent transactions.

Below, you can see how each model predicted that a transaction was fraudulent or valid compared to how often the transaction actually was fraudulent or valid:

According to this data, which model is the best at detecting both fraudulent and valid transactions, and why?