Feature engineering is the critical step that lifts machine-learning models — here's how data transformation improves churn-prediction accuracy in SAP Analytics Cloud.
Building on churn prediction using SAC's Predictive Scenarios, this piece examines feature engineering — a critical machine-learning step — and shows how to apply churn-prediction data within SAC for real-world use cases.
What is feature engineering?
Feature engineering involves selecting, transforming and creating new features from raw data to boost model performance. For classification it enhances the model's ability to accurately categorise data points. The benefits: improved model performance (better accuracy, precision and recall), reduced data noise (filtering out elements that cause overfitting), and dimension reduction through feature selection.
Churn model development
The analysis revealed the 'Age' column affected previous model results by 30.17%. Rather than raw age values, the data was categorised into 'Young', 'Adult' and 'Elderly' groups — and 'Credit Score' received similar treatment. These categorical transformations required adjusting data types from integer to string and statistical types from continuous to nominal.
Model evaluation
Three models were tested: the original approach, one excluding the Age feature and Credit Score group to prevent overtraining, and one removing credit-score information entirely. Model 6 was selected as optimal based on superior Gini scores and overall predictive power.
Conclusion
SAP SAC provides user-friendly interfaces for complex prediction scenarios — classification, time-series analysis and regression — making sophisticated predictive analytics accessible to business users.
