🔮 Demand Forecasting — Supply Chain ML

Supply Chain Analytics Project | Python · scikit-learn · pandas · matplotlib
Part of a logistics optimization portfolio by Emmanuel Beristain Guzmán

---

Overview

Predicts monthly product demand using three progressively advanced models — from a simple moving average baseline to a machine learning ensemble. Built to demonstrate applied knowledge of forecasting methodology, feature engineering, and model evaluation in a supply chain context.

This type of analysis directly supports inventory planning, procurement scheduling, and MRP configuration in systems like SAP.

---

Models Compared

Model	Type	Description
Moving Average	Statistical baseline	Rolling mean of last 3 months
Linear Regression	ML — interpretable	Trend + seasonality features
Random Forest	ML — ensemble	200 trees, lag + calendar features

---

Features Engineered

Feature	Description
t	Time index (captures trend)
sin_month, cos_month	Cyclical encoding of month (captures seasonality)
is_q4	Q4 flag — Oct/Nov/Dec peak season
lag_1, lag_2	Demand from 1 and 2 months ago
lag_12	Demand from same month last year
rolling_3_mean	3-month rolling average
rolling_6_mean	6-month rolling average
rolling_3_std	3-month rolling standard deviation

---

Metrics Used

Metric	Meaning
MAE	Mean Absolute Error — average units off
RMSE	Root Mean Squared Error — penalizes large errors
MAPE	Mean Absolute % Error — model accuracy in %
R²	How much variance is explained by the model

---

Sample Output

══════════════════════════════════════════════════════════
  DEMAND FORECASTING — MODEL COMPARISON REPORT
══════════════════════════════════════════════════════════
  Model                     MAE    RMSE     MAPE      R²
────────────────────────────────────────────────────────
  Moving Average           42.3    51.8    9.87%   0.712
  Linear Regression        28.1    34.5    6.43%   0.881
  Random Forest            18.7    23.2    4.21%   0.947
────────────────────────────────────────────────────────

  Best model: Random Forest (MAPE = 4.21%)

  Top 5 features (Random Forest):
  lag_1                ████████████████ 0.412
  rolling_3_mean       ████████████     0.298
  lag_12               ████████         0.187
  t                    ████             0.063
  sin_month            ██               0.024
══════════════════════════════════════════════════════════

---

Charts Generated

Full forecast chart (4 panels):

• Historical demand + all model predictions
• Zoom into test period (last 6 months)
• MAPE comparison bar chart
• Random Forest feature importance

---

Project Structure

demand-forecasting/
│
├── demand_forecasting.py      # Main script — data, models, charts
├── requirements.txt
├── README.md
│
└── data/                      # Auto-generated on first run
    ├── forecast_chart.png     # 4-panel visualization
    ├── forecast_predictions.csv
    └── model_metrics.csv

---

Getting Started

1. Clone the repo

git clone https://github.com/net421/demand-forecasting.git
cd demand-forecasting

2. Install dependencies

pip install -r requirements.txt

3. Run the forecaster

python demand_forecasting.py

---

How It Works

Raw demand data (36 months)
        ↓
Feature engineering (lag, rolling, calendar)
        ↓
Train / test split (last 6 months = test)
        ↓
Train 3 models simultaneously
        ↓
Evaluate: MAE · RMSE · MAPE · R²
        ↓
Export predictions + chart

---

Business Application

This forecasting logic maps directly to supply chain planning tools:

This script	SAP / ERP equivalent
Moving average	Simple forecast in SAP APO / IBP
Trend + seasonality	Holt-Winters in SAP Demand Planning
Random Forest	Advanced ML in SAP IBP for demand
Feature: lag_12	Seasonal reference in MRP planning
Feature: is_q4	Promotional planning / peak season buffer
MAPE metric	Forecast accuracy KPI in S&OP reviews

Accurate demand forecasting reduces inventory costs, prevents stockouts, and improves supplier scheduling — all core outcomes measured in SAP MM/PP implementations.

---

Tech Stack

Tool	Use
Python 3.10+	Core logic
pandas	Data manipulation
numpy	Numerical computation
scikit-learn	ML models (LinearRegression, RandomForest)
matplotlib	Multi-panel visualization

---

Skills Demonstrated

• Time series feature engineering (lag, rolling, cyclical encoding)
• Supervised ML for regression (Linear, Random Forest)
• Model evaluation and comparison methodology
• Demand planning concepts (MAPE, seasonality, trend)
• Supply chain forecasting aligned with S&OP processes
• Python scripting for logistics automation

---

About

Emmanuel Beristain Guzmán
Logistics Engineer | Supply Chain Analytics | SAP Functional Trainee
📍 México (Remote) · 📧 emmanuel.beristain.guzman@gmail.com
🔗 github.com/net421

---

Part of a series of supply chain analytics projects. See also: EOQ/ROP inventory simulator, logistics dashboard, and VRP route optimization.