Business Results

  • Up to 22% faster clustering in batch

author-image

作者

View All Reference Kits

Background

Losses related to credit card fraud exceeded $12 billion, increasing nearly 11 percent, year over year, in 2021.1 In terms of substantial financial losses, trust, and credibility, this is a concerning issue to banks, customers, and merchants.

E-commerce related fraud has been increasing at a compound annual growth rate (CAGR) of approximately 13 percent.2 Due to the steep increase in fraudulent credit card transactions, it is very important to detect the fraud at the time of the transaction to help consumers as well as banks. Machine learning can play a vital role in predicting fraud by training credit card transaction models, and then using those models to detect fraudulent transactions faster and with a higher degree of accuracy.

This reference kit offers financial institutions the opportunity to implement AI to better predict fraudulent credit card transactions. Banks and financial institutions can lower losses due to credit card fraud by building and training fraud detection models. These models are optimized to run faster and more efficiently and accurately with Intel® software product optimizations integrated into the kit.

Solution

In collaboration with Accenture*, Intel developed this credit card fraud detection AI reference kit. Paired with Intel software products, this kit can help to increase the efficiency, and speed of predicting and detecting fraudulent credit card transactions. This reference kit includes:
 

  • Training data
  • An open source, trained model
  • Libraries
  • User guides
  • Intel® AI software products

At a Glance

  • Industry: Banking, Finance
  • Task:
    • Cluster of similar transactions together followed by classification of credit card transactions as legitimate (class 0) or fraudulent (class 1)
  • Dataset: Credit card transactions
  • Type of Learning: Unsupervised clustering followed by supervised machine learning
  • Models:
    • DBSCAN to cluster transactional data
    • LightGBM model is trained with the clustered data to identify fraudulent transactions.
  • Output: Fraudulent transactions are detected
  • Intel® AI Software Portfolio:
    • Intel® AI Analytics Toolkit (AI Kit)
    • Intel® Extension for Scikit-learn*
    • daal4py (part of Intel® Distribution for Python*)

Since fraud detection can be a compute-intensive operation for inference workloads given the huge scale of credit card transactions in the market, this experiment showcases how the AI Kit speeds up training and inferencing. This kit uses DBSCAN to cluster the transactional data, and then uses the LightGBM model to train and identify fraudulent transactions. Supervised machine learning, using LightGBM, to create increasing datasets of fraudulent transactions. Hyperparameter tuning is then applied for further performance optimization and to enhance prediction accuracy. The trained model inference is calculated in batch with incremental dataset sizes to then benchmark against Intel technology.

Technology

Optimized Intel® AI Software Products for Better Performance.

AI Kit—Achieve end-to-end performance for AI workloads.

Intel Extension for Scikit-learn—Help enable faster and more effective training models. A better trained model can improve the accuracy of purchase predictions.

Performance was tested on Microsoft Azure* Standard_D8_V5 using 3rd generation Intel® Xeon® processors to optimize the kit.

Benefits

Credit card fraud detection can be a compute-intensive operation for inference workloads, given the scale of credit card transactions occurring in the market. This AI reference kit accelerates the data clustering and the inference stages of the solution, which results in less use of compute intensive resources. This helps to greatly reduce an enterprise's total cost of ownership and the time it takes to detect potential fraud. Data scientists can also benefit from accelerating clustering during training to achieve more accurate models, which further enhances fraud detection.

When a credit card is presented for a transaction, merchants and banks must balance between the latency in detecting fraud and completing the transaction quickly. Given that most transactions are valid, delays in reviewing a transaction for fraud could potentially result in the customer abandoning the basket. The additional performance gains enabled by the Intel AI Analytics Toolkit may help reduce fraudulent transactions without incurring too much latency overhead.

Download Kit

Stay Up to Date on AI Workload Optimizations

Sign up to receive hand-curated technical articles, tutorials, developer tools, training opportunities, and more to help you accelerate and optimize your end-to-end AI and data science workflows.

Take a chance and subscribe. You can change your mind at any time.

通过提交此表单,您确认您已年满 18 周岁,并同意就执行此业务请求与英特尔分享您的个人信息。英特尔的网站和通讯受隐私声明使用条款的制约
通过提交此表单,您确认您已年满 18 周岁,并同意就执行此业务请求与英特尔分享您的个人信息。此外,您还同意通过电子邮件和电话订阅来随时了解最新英特尔技术和行业趋势。您可以随时取消订阅。英特尔的网站和通讯受隐私声明使用条款的制约