Supply Chain Optimization using Machine Learning

Overview

This project demonstrates a comprehensive approach to supply chain optimization using machine learning techniques. It covers three main areas:

1. Demand Forecasting: Predicts future product demand using time series analysis and neural networks.
2. Inventory Optimization: Optimizes inventory levels to minimize costs while ensuring product availability.
3. Logistic Optimization: Finds optimal transportation routes using graph-based algorithms and machine learning.

The project is implemented in a Jupyter notebook (optimization_of_supply_chain.ipynb) that walks through data preprocessing, model training, evaluation, and visualization for each optimization area.

Features

• Demand Forecasting:

  • Time series analysis of sales data
  • Multi-layer perceptron (MLP) neural network for prediction
  • Interactive visualizations using Plotly
  • Evaluation metrics including RMSE

• Inventory Optimization:

  • Feature engineering for inventory control
  • Classification model to determine supply actions (Increase/Reduce/No Action)
  • Random Forest classifier implementation
  • Safety stock and optimal stock level calculations

• Logistic Optimization:

  • Graph-based route optimization using OpenStreetMap data
  • Linear regression for distance prediction
  • Route visualization with optimal path highlighting
  • Integration with NetworkX and OSMnx libraries

Data

The project uses two main datasets:

Demand Dataset (demand_dataset/)

• train_0irEZ2H.csv: Training data with weekly sales by SKU and store
• best_nfaJ3J5.csv: Test data for forecasting
• sample_submission_pzljTaX.csv: Sample submission format

Inventory Dataset (inventory_datset/)

• Dynamic Inventory Analytics.xlsx: Excel file containing multiple sheets:
  • Inventory Control: Current inventory levels and parameters
  • SKU Items: Product information
  • Sales Data: Historical sales transactions
  • Warehouse: Warehouse details

Installation and Setup

Prerequisites

• Python 3.7+
• Jupyter Notebook
• Required Python packages (see Dependencies section)

Installation Steps

1. Clone or download this repository
2. Install required dependencies:

   pip install -r requirements.txt

3. Launch Jupyter Notebook:

   jupyter notebook

4. Open optimization_of_supply_chain.ipynb and run the cells sequentially

Dependencies

The following Python packages are required:

pandas
numpy
matplotlib
seaborn
scikit-learn
plotly
chart-studio
keras
tensorflow
osmnx
networkx
openpyxl

You can install all dependencies using:

pip install pandas numpy matplotlib seaborn scikit-learn plotly chart-studio keras tensorflow osmnx networkx openpyxl

Usage

1. Demand Forecasting:

   • Load and preprocess sales data
   • Transform data into supervised learning format
   • Train MLP model for time series prediction
   • Evaluate model performance

2. Inventory Optimization:

   • Merge inventory and sales data
   • Calculate optimal stock levels and safety stock
   • Train classification model for supply recommendations
   • Generate inventory action predictions

3. Logistic Optimization:

   • Load New York City road network
   • Train linear regression model for distance prediction
   • Find optimal routes between locations
   • Visualize routes on interactive maps

Key Results

Demand Forecasting

• MLP model trained with 40 epochs
• Train RMSE: [Value from notebook]
• Validation RMSE: [Value from notebook]
• 29-day lookback window for 90-day ahead forecasting

Inventory Optimization

• Random Forest classifier with 100 estimators
• Classification accuracy: [Value from notebook]
• Confusion matrix and detailed classification report provided

Logistic Optimization

• Linear regression for edge length prediction
• Mean Squared Error: [Value from notebook]
• Optimal route calculation between specified coordinates
• Interactive route visualization

Model Performance

The notebook includes comprehensive evaluation metrics:

• RMSE for regression tasks
• Confusion matrix and classification report for inventory optimization
• Visual comparisons of predicted vs. actual values

License

This project is open-source and available under the MIT License.

Acknowledgments

• Data sources: Provided datasets for demand and inventory analysis
• Libraries: Scikit-learn, TensorFlow/Keras, Plotly, OSMnx, NetworkX
• OpenStreetMap for geographic data