Supply Chain Demand Forecasting & Inventory Decision Support System

Project Overview

This project is an end-to-end supply chain forecasting system that predicts product demand and converts forecasts into operational inventory decisions such as reorder points and stockout/overstock alerts.

The system simulates a real logistics environment with multiple SKUs and locations and provides a dashboard for business users to monitor risks and trends.

Business Problem

Logistics companies must balance:

• Avoiding stockouts (lost sales)
• Avoiding overstock (excess inventory cost)
• Making data-driven replenishment decisions

This project addresses these challenges by:

• Forecasting demand per SKU and location
• Calculating reorder points based on lead time and service level
• Highlighting risk scenarios in a dashboard

Architecture

Data Flow: SQL Database (SQLite)
→ ETL & Feature Engineering
→ Machine Learning Forecast Model (LightGBM)
→ Inventory Decision Logic
→ Streamlit Dashboard

Data

Synthetic but realistic data is generated including:

• Daily sales (orders_daily)
• Daily inventory snapshots (inventory_snapshot_daily)
• Master data (lead time, region, category, unit cost)

Tables:

• orders_daily
• inventory_snapshot_daily
• master_data

Feature Engineering

• Lag features: lag_1, lag_7, lag_14
• Rolling statistics: rolling_mean_7, rolling_std_7
• Calendar features: day_of_week, month

Modeling

Baseline: Naive forecast (lag_1)
Model: LightGBM Regressor

Metrics:

• Baseline MAE ≈ 5.4
• Model MAE ≈ 3.7

Inventory Decision Logic

Reorder Point: mean demand during lead time + safety stock

Stockout Risk: on_hand < forecast demand during lead time

Overstock Risk: on_hand > 1.5 × forecast horizon demand

Dashboard (Streamlit)

Features:

• KPI overview (stockout risk, overstock risk, average inventory)
• Filterable alerts table by region and category
• SKU drill-down view
• Actual vs forecast demand visualization

Dashboard Screenshots

How to Run

Create virtual environment: python -m venv venv
venv\Scripts\activate

Install requirements: pip install -r requirements.txt

Create database schema: python -c "import sqlite3; con=sqlite3.connect('data/sc.db'); con.executescript(open('sql/schema.sql').read()); con.commit(); con.close()"

Generate data: python src/generate_data.py

Build features: python src/etl.py

Train model: python src/train.py

Generate alerts: python src/forecast.py

Run dashboard: streamlit run src/app.py

Technologies

• Python
• SQLite
• Pandas, NumPy
• LightGBM
• Scikit-learn
• Streamlit

Business Value

This system demonstrates how machine learning can be operationalized in supply chain environments to:

• Improve forecasting accuracy
• Support inventory planning decisions
• Provide actionable insights through dashboards

Use Case

Designed as a portfolio project aligned with supply chain analytics and data science roles in logistics companies such as DHL Supply Chain.