Churn Prediction for Mobile Application

Execution

To execute the Dataform project and run the churn prediction pipeline:

1. Run all actions: Use the command dataform run to execute all Dataform actions, including data processing, model training, and prediction.
2. Monitor execution: Monitor the execution in the Dataform web UI or CLI output to ensure all steps complete successfully.

Results and Evaluation

After successful execution, the churn predictions for the test dataset will be available in the user_propensity.best_model_predict view in your BigQuery project.

To evaluate the model's performance:

1. Access predictions: Query the user_propensity.best_model_predict view to retrieve the predicted churn probabilities for each user in the test set.
2. Compare with actual churn: Compare the predictions with the actual churn labels (available in the df_train_eval_test_data view for the "TEST" split) to assess the model's accuracy.
3. Calculate evaluation metrics: Calculate relevant evaluation metrics such as precision, recall, F1-score, and AUC to quantify the model's performance.

Future Improvements

• Feature engineering: Explore additional features or transformations of existing features to potentially improve model accuracy.
• Hyperparameter tuning: Conduct more extensive hyperparameter tuning for the machine learning models to optimize their performance.
• Model comparison: Evaluate and compare other machine learning algorithms beyond those currently implemented.
• Real-time prediction: Investigate methods for generating real-time churn predictions for new users as they interact with the application.
• A/B testing: Design and conduct A/B tests to measure the impact of interventions based on churn predictions.