Artificial Neural Networks for Fraud Detection in Supply Chain Analytics: A Study on MLPClassifier and Keras
In this study, we aimed to detect fraudulent activities in the supply chain through the use of neural networks. The study focused on building two machine learning models using the MLPClassifier algorithm from the scikit-learn library and a custom neural network using the Keras library in Python. Both models were trained and tested on the DataCo Supply Chain dataset. The results showed that the custom neural network achieved an accuracy of 97.67% in detecting fraudulent transactions, demonstrating its potential to minimize financial losses for organizations.
- Python
- SciKit Learn
- Seaborn
- Pandas
- TensorFlow
- Keras
- Matplotlib
- Plotly
In recent years, the use of neural networks in supply chain analytics has gained considerable traction as organisations look for ways to improve their operations and make more informed decisions. One area where neural networks can have a significant impact is in the detection of fraud before shipments are processed. Fraudulent activities can cause significant financial losses, and early detection is essential for minimising any damage. We present our study on the use of neural networks for detecting fraud in the supply chain. Two models were developed as part of this study: one using the MLPClassifier algorithm from the scikit-learn library, and another using a custom neural network built using the Keras library in Python. These models were developed using open-source libraries, including NumPy for numerical computation, Pandas for data manipulation, Seaborn for statistical data visualisation, matplotlib for plotting, and the machine learning frameworks SciKit Learn, Keras, and Tensorflow (backend). The scikit-learn library is a widely used machine learning library in Python, and the MLPClassifier algorithm from this library is a type of multi-layer perceptron classifier that has been shown to perform well on various classification tasks. The custom neural network, on the other hand, was designed to provide a deeper level of control over the architecture and training process, allowing for a more customised solution. The objective of our study is to identify potential fraudulent activities in the supply chain before shipments are processed, thus reducing the risk of financial loss for the organisation.
