Solved! Go to Solution.
Hi @TobeT ,
To add onto Emily's response, one other area to explore is "Prediction Explanations". This will show you at the per row level (e.g. customer) what features contribute to that particular model's predictions and how strong they are. So, for example, you might see that for Customer 123 they are more likely to have a phone repair because they have replaced 3 phones in the past year, are 17 years old, etc.
With regards to trends, are you using DataRobot's time series product, AutoTS? If so that product will automatically generate hundreds to thousands of features that capture trends (e.g. rolling 7 day, 14 day, 28 day averages, seasonal trends, etc.). These are "special patterns" will be incorporated into the models and can be visualized in the Feature Impact and Feature Effects screens for all models.
Does that help?
Yes, DataRobot is actually quite good at picking up patterns in your data. One place that you can find your answer is under the Understand >> Feature Effects (see below).
Here you can find charts that look at the relationship between the feature and the target variable using a partial dependence metric. This is a model agnostic approach, so you can get this for every one of you models in the platform. Here is a link to more information about interpretability tools.
In this specific example for a readmission use case, you can see that as inpatient visits increases, the likelihood to be readmitted to the hospital also increases.
I hope this answers your question, and thanks for posting again!