Techjays Forecasting & Inventory Optimization – Production‑Grade ML Pipeline

Executive Summary

This project delivers a production‑grade forecasting and inventory optimization engine for 80 SKUs, built using a robust ML pipeline. The system processes multi‑sheet operational data, generates 90‑day demand forecasts, computes inventory control metrics, and automates best‑model selection per SKU — enabling data‑driven procurement and stock management.

Tech Stack

• Python, Pandas
• ARIMA, XGBoost
• Rolling Origin Cross‑Validation
• Power BI

Key Achievements

• Processed 10,000+ rows across Sales, Inventory & Receiving datasets.
• Engineered 30+ features (lags, rolling windows, volatility, calendar, outlier flags).
• Evaluated 6 forecasting models: MA, SES, Holt, Holt‑Winters, ARIMA, XGBoost.
• Implemented Rolling Origin Cross‑Validation for real‑world robustness.
• Automated best model selection per SKU.
• Generated 90‑day forecasts for all SKUs.
• Computed Safety Stock, Lead Time Demand, Reorder Point.
• Delivered inventory‑ready outputs & business recommendations.

Impact

• Enabled forecast‑driven inventory planning.
• Reduced stockout risk across critical SKUs.
• Improved procurement accuracy and operational efficiency.
• Delivered a scalable ML pipeline for production environments.

Key Insight Summary

The pipeline combines classical forecasting, machine learning, and inventory science to deliver actionable SKU‑level predictions. Rolling Origin CV ensured stability, while automated model selection maximized accuracy across diverse demand patterns.

View Project ← Back to Home