Model Explainability with SHAP
Dual-algorithm explainability infrastructure: SHAP for tree models, coefficient contribution for linear models, with BigQuery logging.
Akash Sharma
Primary Owner
Dual-Algorithm
Explainability Coverage
The Problem
Revenue leaders and customers couldn't understand why the forecast changed week-over-week. Black-box outputs eroded trust in the model.
What Was Built
Implemented dual-algorithm explainability: (1) SHAP TreeExplainer for XGBoost models with base value decomposition and Shapley value packaging. (2) Coefficient-based contribution for Ridge/linear models. Both produce human-readable feature display name mappings and marketing-vs-baseline contribution aggregations. All explanations logged to BigQuery for historical tracking and customer-facing dashboard widgets.
Business Impact
Made forecast outputs interpretable to non-technical revenue leaders, improving stakeholder trust and enabling data-driven forecast discussions.
Related Revenue Forecasting Projects
Revenue Forecasting Platform
Production ML system predicting pipeline and booking outcomes at daily frequency across current and future quarters.
Time-Decay Forecast Smoothing
Dynamic adjustment layer that re-weights prior EOQ history by recency, capturing momentum and velocity signals closer to quarter end.
Seasonal Index Forecast Adjustment
Statistical fallback layer using day-of-quarter index averages when ML model signals are insufficient.