Stock Out/Risk Prediction for Inventory Optimization

Tech Stack: Python, SQL, Pandas, Scikit-learn, SHAP, Dataiku DSS, Git

• Led the development of stock risk/out models that generated preemptive alerts, savings approx. $500k per stock out prevention.
• Designed a framework to capture time-based patterns by generating multiple time-aggregated features.
• Developed an ensemble of LightGBM and XGBoost models, achieving an AUROC of 0.87 and Recall of 0.76. This outperformed the client's internal projection system, which had an AUROC of 0.51.
• Used SHAP to identify top global and local predictors of stock risk and stock out.