In other cases, running a pipeline gives full customization of the workflow and model testing by including steps for data cleaning and data splitting. Pipelines also can help the user specify which machine learning algorithm to use. While the highest possible accuracy is usually desired, this lets the user pick algorithms that are less resource-intensive if costs are a factor.
In this case, running a pipeline with various classification algorithms granted passable accuracy, but not as high as the best AutoML run. Since the data required little prep or cleaning for this experiment, we deployed the best AutoML model.
Azure ML also provides an explanation feature that ranks the importance of the features from the dataset. Below we see Age, Balance, and Credit Score are the most important features/indicators for customer attrition in this case, so we can focus on those in the dashboard.