📦 Predictive Supply Chain Optimization

This repository hosts an end-to-end data science pipeline designed to solve one of the most expensive problems in logistics: Inventory Volatility.

Using the Olist E-commerce dataset, this project moves beyond traditional "static averages" to implement a probabilistic inventory model driven by Machine Learning (Facebook Prophet) and statistical risk management.

Now features an fully automated pipeline and an interactive executive dashboard.

---

🚀 Quick Start

1. Run the Pipeline

Execute the full data engineering, analysis, and optimization pipeline with one command:

python run_pipeline.py

This script will:

1. Process raw data.
2. Train forecasting models.
3. Optimize inventory parameters.
4. Generate static figure assets in results/figures/.

2. Launch the Dashboard

View the results in an interactive web application:

streamlit run dashboard.py

This allows you to verify the impact of different service levels and costs on your bottom line in real-time.

---

📂 Repository Structure

File	Description
run_pipeline.py	Automation Orchestrator: Runs all notebooks sequentially to generate results.
dashboard.py	Interactive App: Streamlit dashboard for stakeholders to explore scenarios.
notebooks/01_Data_Engineering.ipynb	ETL Pipeline: Merges 9 raw relational tables into a clean Master Source of Truth.
notebooks/02_ABC_Analysis.ipynb	Strategic Segmentation: Uses Pareto Principle (80/20) to identify high-value SKUs.
notebooks/03_Forecasting.ipynb	Demand Forecasting: Trains a Prophet model to predict weekly sales trends.
notebooks/04_Inventory_Optimization.ipynb	Risk Modeling: Transforms forecasts into actionable Reorder Points (ROP).
notebooks/05_Executive_Report.ipynb	Simulation & ROI: 90-day "Sawtooth" simulation to stress-test the policy.
results/figures/	Generated Assets: Automatically generated plots (Cost Comparison, Sawtooth Chart, etc.).

---

🧠 Methodology: The "Stochastic" Approach

Instead of guessing, we use the Root Sum of Squared Error formula to account for independent variabilities in Supply and Demand.

$$ Safety Stock = Z \times \sqrt{(\bar{L} \times \sigma_D^2) + (\bar{D}^2 \times \sigma_L^2)} $$

• $Z$ (Service Factor): We use 1.65 to target a 95% Service Level (statistically guaranteeing stock availability 95% of the time).
• $\sigma_L$ (Supply Risk): Measures how unreliable the supplier is (Standard Deviation of Lead Time).
• $\sigma_D$ (Demand Risk): Measures how volatile customer purchasing is (Standard Deviation of Daily Sales).

Key Insight: By decoupling these risks, we can hold less inventory for stable products and more for volatile ones, optimizing total working capital.

---

📊 Results & Impact

The 90-day simulation stress-tested this policy against a reactive baseline:

• Service Level Stability: Maintained >95% availability even during simulated demand spikes.
• Capital Efficiency: Reduced holding costs for Class C (low value) items by strictly limiting their safety stock.
• Operational Automation: Generated automated "Reorder Point" triggers, removing manual guesswork.

---

🛠️ Installation & Setup

1. Prerequisites

• Python 3.11+
• Anaconda or Miniconda

2. Environment Setup

It is recommended to use a separate Conda environment to ensure library compatibility.

# Create the environment
conda create -n supply_chain_311 python=3.11

# Activate it
conda activate supply_chain_311

# Install dependencies
pip install -r requirements.txt

---

⚖️ License

Code: The software logic in this repository is licensed under the MIT License.

Data: The dataset is provided by Olist under CC BY-NC-SA 4.0.