User churn prediction-Platform For AI(PAI)-阿里云帮助中心

Use PAI to quickly build a user churn prediction model.

Background

Building a user churn prediction model allows you to identify customers who are likely to churn and take proactive measures to retain them.

Traditional churn monitoring solutions are often rule-based. These methods lack intelligent prediction capabilities and may fail to accurately identify at-risk users.

Solution

PAI provides a complete solution for feature encoding, classification model training, and model evaluation based on labeled data. The requirements are as follows:

Expertise: Basic knowledge of machine learning modeling.
Development cycle: 1 to 2 days.
Data: A labeled dataset with thousands of records describing the characteristics of customers who have churned.

Dataset

This pipeline uses an anonymized dataset of real customer behavior from the telecommunications sector. It contains 7,043 samples, which include basic user information and churn attributes. The feature data includes the following fields.

Parameter	Description
customerid	The customer ID.
gender	The customer's gender.
SeniorCitizen	Indicates whether the customer is a senior citizen. Valid values: 1: Yes. 0: No.
Partner	Indicates whether the customer has a partner.
Dependents	Indicates whether the customer has dependents.
tenure	The number of months the customer has been with the company.
PhoneService	Indicates whether the customer subscribes to a phone service.
MultipleLines	Indicates whether the customer has multiple lines.
InternetService	The customer's internet service provider. For example, DSL or Fiber optic.
OnlineSecurity	Indicates whether the customer subscribes to the online security service.
OnlineBackup	Indicates whether the customer uses the online backup service.
DeviceProtection	Indicates whether the customer has the device protection service.
TechSupport	Indicates whether the customer has requested technical support.
StreamingTV	Indicates whether the customer subscribes to the streaming TV service.
StreamingMovies	Indicates whether the customer subscribes to the streaming movies service.
Contract	The contract term. For example, Month-to-month or Two-year.
PaperlessBilling	Indicates whether the customer uses paperless billing.
PaymentMethod	The payment method.
MonthlyCharges	The monthly charge.
TotalCharges	The total charges.

The target data includes the following field.

Parameter	Description
churn	Indicates whether the customer has churned.

Procedure

Go to the Machine Learning Designer page.
1. Log on to the PAI console.
2. In the left-side navigation pane, click Workspaces. On the Workspaces page, click the name of the workspace that you want to manage.
3. In the left-side navigation pane, choose Model Training > Visualized Modeling (Designer).

Build the pipeline.

On the Designer page, click the Preset Templates tab.
In the Churn User Monitoring section, click Create.
In the Create Pipeline dialog box, configure the parameters. You can use the default settings for all parameters.

The Data Storage parameter specifies an OSS Bucket path to store temporary data and models generated during the pipeline run.
Click Confirm.

It takes about 10 seconds to create the pipeline.
In the pipeline list, double-click the Churn User Monitoring pipeline to open it on the canvas.

The system automatically builds the pipeline based on the preset template, as shown in the following figure.

用户流失工作流

Section	Description
①	The dataset for the pipeline.
②	You can use the One Hot Encoding and SQL Script components to perform feature engineering and convert original string-type features into numerical features. For example, for the target field churn, which has the original values Yes or No, you can use an SQL statement to convert Yes to 1 and No to 0, as shown in the following example. `select (case churn when 'Yes' then 1 else 0 end) as churn from ${t1};`
③	The data is split into a training dataset and a prediction dataset. Because a customer can either churn or not churn, user churn prediction is a binary classification problem that can be solved with binary classification algorithms.
④	The Binary Classification Evaluation component validates the model. You can evaluate its performance using metrics such as AUC, KS value, and F1 score.

Run the pipeline and view the model results.
1. At the top of the canvas, click Run.
2. When the pipeline run is complete, right-click the Binary Classification Evaluation component on the canvas and select Visual Analysis from the shortcut menu.
3. In the Binary Classification Evaluation dialog box, click the Indicator Data tab to view the model evaluation metrics.
  
  The metrics include KS value, AUC, F1 score, Negative Samples, Positive Samples, and Total Samples. An AUC value closer to 1 indicates higher prediction accuracy. In this example, the AUC value is greater than 0.8, indicating high model prediction accuracy.