Fraud Detection in Supply Chains Using Kolmogorov Arnold Networks

This project, titled "Enhancing Fraud Detection in Supply Chains with Kolmogorov Arnold Networks: A Comparative Analysis with Multi-Layer Perceptrons (Initial Findings and Methodology)," aims to enhance fraud detection in supply chains by leveraging Kolmogorov Arnold Networks (KANs) and contrasting their performance with traditional Multi-Layer Perceptrons (MLPs).

Project Overview

The study explores the application of Kolmogorov Arnold Networks (KANs) in detecting fraud within supply chains, emphasizing their interpretability and adaptability compared to traditional Multi-Layer Perceptrons (MLPs). KANs utilize learnable activation functions, enhancing model transparency and enabling more effective fraud detection.

Repository Structure

• Article: Contains the detailed study and findings.
• Classification: Scripts and models related to the classification tasks.
• PCA_and_other_approaches: Alternative approaches and dimensionality reduction techniques.
• Regression: Scripts and models related to regression tasks.
• DataCoSupplyChainDataset.zip: The dataset used for the study.
• DescriptionDataCoSupplyChain.csv: Description of the dataset features.
• README.md: This file.

Data Collection

The dataset comprises transactional data from a supply chain, including features such as transaction amounts, product categories, customer information, and timestamps. The dataset can be found here.

Data Processing

Data Cleaning

• Removed unnecessary columns and rows with missing values in critical fields.
• Combined customer names into a single column.
• Split order dates into separate year, month, day, and hour columns.

Feature Engineering

• Created a target variable named 'fraud' based on the 'Order Status' and 'Type' category.
• Unified first and last names into a 'Cust_Full_Name' column.
• Formatted date columns for consistency and time-based analysis.

Encoding Categorical Variables

• Identified and label-encoded categorical columns to convert them into numerical values.

Handling Class Imbalance

• Applied the Adaptive Synthetic (ADASYN) algorithm to balance the training data to address the imbalance in the data with lower fraud values.

Converting Data to Tensors

• Split the dataset into training and test sets while stratifying the training data to ensure a balance in the dataset.
• Converted the preprocessed data into PyTorch tensors.

Model Development

KAN Setup & Layer Configuration

• Constructed using the KAN library with learnable activation functions.
• Configured with an input layer, six hidden neurons, and output neurons for classification tasks.

Training Process

• Utilized the LBFGS optimizer for high-dimensional problems.
• Trained for 5 epochs, tracking both training and test accuracy.
• Saved visual representations of the training process.

Evaluation Metrics

• Derived a symbolic formula for the decision boundary with train and test accuracies of 90.44% and 99% for formula 1 and train and test accuracy of 89.53% and 98% respectively for formula 2. (Selected Formula 1 based on the score for this reason)
• Visualized model activations and pruned less significant connections for efficiency.

Analysis and Insights

• Accuracy: Substantial improvement in test accuracy with a smaller dataset seen with the test and train accuracy.(recommended training for higher steps)
• Formula Derivation: Captured underlying patterns of fraudulent transactions based on each activation function present on the nodes.
• Visualization and Pruning: We then plot each activation function that can be analyzed further and then prune the model to ensure an efficient model performance while maintaining accuracy.

Suggestions

• Compute Resources: Larger datasets require greater compute power, suggesting the need to build the model on the GPU for better compute.
• Ideal Applications: Suitable for various fields including pharmaceutical, manufacturing, market research, energy sector, environmental science, and legal compliance.

Requirements

The project requires the following dependencies:

Requirements

# python==3.9.7
matplotlib==3.6.2
numpy==1.24.4
scikit_learn==1.1.3
setuptools==65.5.0
sympy==1.11.1
torch==2.2.2
tqdm==4.66.2

For installation of these packages, please refer to the installation guide.

Contributing

Contributions to the project are welcome. Please fork the repository, make your changes, and submit a pull request. For major changes, open an issue first to discuss what you would like to change.

References

• Soori, M., & Arezoo, B. (2023). Artificial Neural Networks (ANNs) in supply chain management: Opportunities and challenges. Journal of Economy and Technology, 18(3), 87-102. DOI
• Das, S. (2023). Artificial Neural Networks for Fraud Detection in Supply Chain Analytics: MLPClassifier and Keras. GitHub repository
• Liu, Z., et al. (2024). Kolmogorov-Arnold Networks. arXiv preprint

For more details, refer to the Article directory.

Citation

@article{liu2024kan,
  title={KAN: Kolmogorov-Arnold Networks},
  author={Liu, Ziming and Wang, Yixuan and Vaidya, Sachin and Ruehle, Fabian and Halverson, James and Solja{\v{c}}i{\'c}, Marin and Hou, Thomas Y and Tegmark, Max},
  journal={arXiv preprint arXiv:2404.19756},
  year={2024}
}

To Cite me

@project{ChrisD-7,
  title={Enhancing Fraud Detection in Supply Chains with Kolmogorov Arnold Networks: A Comparative Analysis with Multi-Layer Perceptrons (Initial Findings and Methodology)},
  author={Chris DSilva},
  year={2024},
  howpublished={\url{https://github.com/ChrisD-7/Fraud-Detection-in-Supply-Chains-with-Kolmogorov-Arnold-Networks/blob/main/README.md}}
}

Classification Model

EDA and Model Building with Pruning

Activation Function Plots