Another fundamental phase of the CRISP-DM methodology is the evaluation phase, which focuses on the quality of the model, and its ability to meet the overall business objectives. If the model can't meet the objectives, then it's important to understand if there is a business reason why the model doesn't meet the objectives, in addition to technical possibilities that might account for failure. It's also a good time to pause and consider the testing results that you have generated thus far. This is a crucial stage because it can reveal challenges that didn't appear before. That said, it is an interesting phase because you can find new and interesting things for future research directions. Therefore, it's important not to skip it!

Fortunately, we can visualize the results using Tableau so that the neural networks are easier to understand. There are several performance measures for neural networks, and we will explore these in more detail along with a discussion of how the performance measures are visualized. You can view the results as Receiver Operator Characteristic (ROC) curves, Precision/Recall curves, or Lift curves. Additional data visualizations could include a confusion matrix, and cumulative values for the area under the curve (AUC). Let's look at these measurements in more detail.