The project

The customer, a leading bank offering financial services such as deposits, loans, and more, hired our partner’s team to develop a cutting-edge machine learning solution to detect and combat financial fraud.

To address this challenge, the team developed an ML-based fraud detection extension, utilizing deep learning algorithms to identify anomalies in banking transactions.

Our experts collected and consolidated vast amounts of bank-related data, uncovering unique patterns—such as unusually high-value or fragmented transactions—that help machine learning models detect fraudulent activity.

They leveraged a large dataset containing millions of samples, including neural network data, transactional history, and feature selection techniques, to train the ML algorithms. This approach enables the system to recognize unknown or high-risk transactions, even with limited data.

When a potential threat is detected, the system immediately alerts the administrator, allowing them to pause or cancel suspicious transactions for further investigation, ensuring proactive fraud prevention.

Technologies

Python, Scala, Apache Spark MLLib, Scikit-learn, LightGBM, XGBoost, Hyperopt, PySpark, Numpy, Pandas, Scipy, Apache Flink,

Redis Feast, Apache Hive, Apache Airflow, Apache Kafka, Apache Spark, React, HTML, CSS

Team

DDT, 1 developer, 2 ML engineers