Single Data Pipeline Powers Both BI Dashboard and ML Models, Preventing Data Drift

A data engineer built a unified governed pipeline producing one master dataset — adult_master.csv — from the UCI Adult Census Income data, serving both a Power BI dashboard and four machine learning models simultaneously. The approach was chosen to prevent data drift, a common problem when separate datasets are maintained for analytics and modeling teams. Missing values were preserved as 'Unknown' with boolean tracking flags rather than being silently imputed, keeping the pipeline fully auditable. Permutation importance tested across all four models revealed that marital status, capital gains, and education were the strongest predictors of income bracket. Among the models evaluated, XGBoost achieved the highest ROC-AUC score of 0.928, narrowly outperforming HistGradientBoosting at 0.927.