This article explains the type of information that is in the overview of an created using the customer churn prediction template.
Once the user creates an app using the Customer Churn Prediction template, they will be directed to the app overview page. The information regarding the app's setup is divided into three sections on this page. The particulars are outlined as follows:
1. App setup
The app setup provides a comprehensive overview of two essential aspects: the data processing flow and the prediction target.
The data processing flow describes how the input datasets are prepared, and transformed into data for model training. The prediction target indicates the kind of prediction that the model will make based on the business option selected. The model can either predict the probability if customer churn in a future time frame, based on their transaction patterns (for the transactional option) or predict the urgency of applying retention measures, based on their historical behavior (for the subscription option).
In the overview, users can see the high level app setup:
App setup of the transactional option
App setup of the subscription option
A visual representation of the data-preparation pipeline generated by the template is also available (“View more” under the “Data preparation” section):
Example of data-preparation pipeline for the transactional option
Example of data-preparation pipeline for the subscription option
2. Making predictions
This section provides the following information:
The best model and its prediction quality, if the training is complete
The number of models in training, in evaluation or ready
A button to quickly make a prediction using the available model(s)
For more information on prediction quality, read this article.
3. Prediction results
The app overview page also displays the results of the predictions made by the users. Depending on the mode of prediction, a result may consist of multiple files. The users have the option to select either a one-time prediction or a periodic prediction, which generates one file per prediction at regular intervals. The prediction view lists the following information fields:
A preview of the data and the predictions, which provides a brief overview of the outcome
A details page, which can be accessed by opening the “View details” link, that explains how the prediction result was generated, demonstrating the underlying process
A download option, which enables saving the prediction result to the device, supporting further analysis or backup
An export function, which facilitates transferring the prediction result to another location, such as a database or a project on the Engine, for integration or collaboration
A delete action, which permits removing the prediction result from the app