Building a Churn Predictor

Building a Churn Predictor with Snowflake

This workflow connects to a Snowflake database, loads and joins customer data, computes summary statistics, and prepares the data for machine learning. The data is split into training and test sets. The training data is converted for use with H2O, and a Random Forest model is trained to predict customer churn. The trained model is saved for future use and also applied at scale to the test data directly in the database. Finally, the workflow evaluates the model's performance using accuracy statistics and a ROC curve to measure how well churn is predicted.

Connect to Snowflake database

Write Telco data to Snowflake and join tables

Compute in-database summary stats

Sample and partition the dataset

Convert to H2O context, train and apply ML model

Score ML model

📍Technical Note:

If you don't have a Snowflake account, you can sign up for a free 30-day trial account (when signing up, select Enterprise edition and choose any Snowflake cloud/region - preferably AWS or Azure).

Configure the Snowflake Connector node:

After signing up, navigate to your Snowflake Account details. There, you'll find key information to configure the node:

• Account identifier (input it in Full account name field)

• Login name (input it in Authentication > Username & password field). The password field is the password used to log in to your Snowflake account.

• Role (input it in Default access control role). Make sure the role is ACCOUNTADMIN.

• Stay in your Account details and navigate to the Config File tab. Select the warehouse you prefer (e.g., SNOWFLAKE_LEARNING_WH, if you're using a free trial account) and input the warehouse name in the Virtual warehouse field.

📍Technical Note:

[For users with a free 30-day trial account] In the Write Telco data to DB component, keep the default configurations and simply execute the component. The Table names are defined automatically.

[For users with a regular account] Double-click on the Write Telco data to DB component and provide your preferred Database and Schema names (they must already exist). The Table names are defined automatically.

💡Pro tip: If your working with very large datasets, we recommend replacing the DB Writer node with the DB Table Structure Creator + DB Loader nodes for faster, bulk data loading to database.