Data Used

We will use data on the service usage of active Netflix subscribers who have not yet canceled.

This dataset includes customer ID, First use date, Last use date, content viewing status, and customer cancelations. Since the data is for active subscribers, the Cancel flag is “FALSE” for all customers.

Running “Predict with Model (Analytics View)”

To predict future cancellation risks for active customers, we will use a survival analysis model previously created in a different data frame.

Click the step menu of your target data frame and select “Predict with Model (Analytics View).”

In the model selection dialog, specify the data frame where your model was created. Here, we select the “Customer usage status” data frame and then choose the pre-built survival analysis model from the Analytics Model list.

Survival Rate After N Months

First, let’s look at “Survival Rate after N months,” which predicts the survival rate after a specified period.

Check the “Survival Rate after N months” option under prediction type.

Next, set the baseline date. Select “Specific Time” and choose the baseline date from the calendar picker. In this case, we’ll use January 31, 2023, which is the data acquisition date or the last date in the dataset.

You can also select “Today” as the baseline date.

Since we want to calculate the survival probability one month from the baseline date, enter “1” in the prediction period and click the “Run” button.

Reviewing and Interpreting Prediction Results

After the prediction is complete, a new prediction step is added, along with several orange-highlighted columns.

The added columns include:

• Base Time: The specified baseline date.
• Base Survival Time: Survival period up to the baseline date.
• Prediction Time: The target prediction date.
• Prediction Survival Time: Survival period at the prediction point.
• Predicted Survival Rate: The predicted survival rate.
• Linear Predictor: A risk score. Higher values indicate a higher relative event rate and a lower survival rate.
• Std Error: The standard error of the Linear Predictor.

Interpreting the Predicted Survival Rate

It’s important to note that the Predicted Survival Rate represents a conditional survival rate from the baseline date, not an absolute survival rate.

The prediction here is the survival rate from the baseline date, not the survival rate “until the specified date.”

For example, if James started using the service in January 2023, and his survival rate one month after the baseline date (January 31, 2023) is predicted to be 60%, this means an absolute survival rate of 60% one month after service start (a change from 100% to 60%).

On the other hand, if Nancy started in September 2022 and has already been a customer for four months by the baseline date, his survival rate one month later is predicted to be 91%, assuming 100% survival at the baseline. This represents a change from 35% to 32% on the survival curve (32% / 35% = 91%).

Based on this interpretation, James’s first month survival rate is 40%, while Nancy’s is 91%, indicating James is at higher risk.

Thus, it’s crucial to understand that “Predicted Survival Rate” refers to the “survival rate from the baseline date.”

Prediction Limitations

When reviewing prediction results, you might see specific messages in the note column.