Post by Younes Abdelkefi
Data Engineering student | AI & ML Enthusiast | IEEE Volunteer
š End-Of-Year Project ā Real-Time E-Commerce Analytics & Fraud Detection Platform. š³ In today's digital economy, millions of transactions generate valuable data every day. The challenge lies in analyzing this data in real time to support business decisions and identify suspicious activities. This challenge inspired our End-Of-Year Project: "Real-Time E-Commerce Analytics & Fraud Detection Platform", a scalable system that combines Big Data technologies, stream processing, graph-based machine learning, and real-time analytics to monitor transactions and identify fraudulent activities as they occur. In today's digital economy, e-commerce platforms generate massive streams of transactional data every second. Detecting suspicious behavior in real time requires not only accurate machine learning models but also a robust and scalable data infrastructure capable of processing high-throughput event streams with minimal latency. āļø Key Technical Highlights š¹ Designed and deployed a real-time streaming architecture capable of handling 50,000+ events per hour with sub-200 ms latency. š¹ Built a scalable data pipeline using distributed stream-processing technologies for continuous ingestion, transformation, and monitoring of transaction data. š¹ Leveraged ClickHouse for high-performance analytical storage, reducing dashboard query times by 30% through optimized aggregation strategies. š¹ Implemented real-time analytics and aggregation workflows to provide up-to-date business and operational insights. š¹ Integrated a Graph Neural Network (GAT) fraud detection model on transaction graphs to capture complex relationships between users, transactions, and suspicious activities. š¹ Developed interactive Power BI dashboards featuring KPIs, anomaly alerts, fraud monitoring indicators, and drill-down analytics for business users. š¹ Ensured scalability, fault tolerance, and low-latency performance to support large-scale transaction processing. š Results ā Real-time processing of high-volume transaction streams with fast analytical querying and monitoring capabilities. ā Fraud detection powered by Graph Neural Networks, achieving a 92% F1-score on transaction graph data while maintaining a low false-positive rate. ā Improved operational visibility through real-time dashboards and intelligent alerting mechanisms. š„ Team This project was developed in collaboration with Khalil Zaghla. š Acknowledgments I am sincerely grateful to our supervisor, Mrs. Ines Ben Messaoud, for her guidance and continuous support throughout this project, and to our examiner, Mrs. Molka Rekik, for her valuable feedback and insightful recommendations. I look forward to applying these learnings to future AI- and data-driven challenges. š GitHub Repository https://lnkd.in/dMSidHfn #BigData #DataEngineering #Streaming #ClickHouse #MachineLearning
