CS613: Machine-Learning

Demand Forecasting

About The Project

The main aim of this project is to make sure each shop receives the right amount of stock at the right time. It focuses on avoiding situations where a shop runs out of products and loses customers, while also preventing sending more stock than the shop can realistically sell. In simple terms, the objective is to maintain a healthy balance—not too much, not too little—so that shops can operate smoothly, meet customer demand, and avoid unnecessary costs or wastage.

Built With

This project is developed using the following technologies:

• Python
• Pandas
• NumPy
• Scikit-learn
• LightGBM
• XGBoost
• Matplotlib & Seaborn
• Jupyter Notebook

Getting Started

To get a local copy up and running follow these steps.

Prerequisites

Before you begin, ensure you have the following installed:

• Python 3.10 or higher
• Jupyter Notebook / JupyterLab
• pip (Python package installer)

To install Jupyter Notebook:

  pip install notebook

To verify Python version:

  python --version

Installation

Follow these steps to run the project on your machine:

1. Clone the repository

   git clone https://github.com/khushiipatel/FinalProjectML.git 

2. Navigate into the project

   cd repo_name

3. Open Jupyter Notebook

   jupyter notebook 

4. Open the project file Inside Jupyter, open:

   Project.ipynb 

5. Run all cells

Dataset

• The dataset is taken from the Google Cloud public Marketplace in Big Query.​
• Sales Data: Daily sales transactions​
  • Fields: sales_date, dealer_code, product, qty, total_price​
• Stock Data: Daily inventory snapshots​
  • Fields: benchmark_date, ec_id, product_name, inventory_qty​
• Scale: Multiple stores × multiple products × daily timestamps​
• Preprocessing: Merged datasets, handled missing values, created time-series structure.

Visualization

This visualization helps identify which model most accurately predicts daily demand across all products for each of the 7 days, enabling better decision‑making for inventory and resource planning.

For detailed model experiments, evaluation metrics, and performance comparisons, please refer to the Documentation

System Architecture

Machine Learning Models used

• LightGBM;
• Support Vector Regression (SVR);
• Complementary Models: XGBoost & Random Forest;
• Ensemble Strategy

Results

Model Performance

Model	RMSE	MAE	WMAPE
LightGBM	0.9573	0.3531	1.09%
XGBoost	1.0774	0.4227	1.30%
Random Forest	2.3733	1.2565	3.86%
SVM(RBF)	12.0908	6.1007	18.76%
ESEMBLE	1.0383	0.4559	1.40%

LightGBM achieved the best overall performance with the lowest RMSE and WMAPE.

Output

Example Use Cases

This demand forecasting system helps businesses make better inventory decisions:

• Reduce missed sales from stockouts by identifying which products are likely to run out and need restocking soon.
• Minimize waste from overstock by detecting products with excess inventory before they become obsolete.
• Optimize inventory costs by planning purchasing cycles to keep just enough stock without tying up too much capital.
• Improve customer satisfaction by ensuring popular products are always available when customers want to buy them.

References

H. Malik, "A beginner's approach to time-series with working example: Demand forecasting │ Time series example with Kaggle," Medium, Jun. 11, 2024. [Online]. Available: https://medium.com/@humzahmalik/a-beginners-approach-to-time-series-with-working-example-c6bff9c24928

License

This project was completed as part of academic coursework for CS613: Machine Learning.