Optimizing Ad Conversions with DS / Yael Kiselman (DigitalTurbine)

Optimize ad conversions using data science techniques like XGBoost, Optuna, and SHAP, and learn how to deploy models in production while addressing the cold start problem.

Key takeaways
  • Optimize Ad Conversions: Optimize ad conversions by predicting user behavior and tailoring models to individual users.
  • Use Lag Features: Use lag features, such as time series data, to improve model performance and capture sequential dependencies.
  • XGBoost: Use XGBoost, a powerful machine learning algorithm, to handle large datasets and optimize conversion rates.
  • Optuna: Use Optuna, a Bayesian optimization library, to search for the best hyperparameters and optimize model performance.
  • Pandas and Spark: Use Pandas and Spark for data processing and analysis, and leverage PySpark for big data processing.
  • Grouping Users: Group users based on similarity or features to improve model performance and reduce data complexity.
  • Lift and Conversion Rate: Measure lift and conversion rate to evaluate model performance and optimize ad campaigns.
  • SHAP: Use SHAP, a machine learning explainability library, to interpret model predictions and understand feature importance.
  • Data Engineering: Perform data engineering tasks, such as data preprocessing and feature engineering, to prepare data for modeling.
  • Model Deployment: Deploy models in production and monitor performance to identify areas for improvement.
  • Cold Start Problem: Address the cold start problem by using techniques such as preloading apps and using historical data to improve model performance.
  • Real-time Processing: Leverage real-time processing capabilities to enable fast and efficient data processing and analysis.