Developing Data Science Framework, Architecture and Methodology for Fraud Detection in ATM Transactions in Banks
Financial institutions are adopting new and enhanced monitoring practices and technologies to prevent and detect increasingly common occurrences of fraudulent financial transactions of ATM, Credit card and Internet banking. The preventive or corrective measures include better tampering detection and notification, analytic solutions that identify anomalies in the transactions such as detecting a fraudulent ATM transaction to withdraw cash. Fraud detection/prevention is very complex and it can be characterized by the fact that they may use unstructured data or only structured data, they require real-time as well as batch processing, use a rules engine or derive a pattern or sequence. The key challenges in fraud detection are: There are no universal fraud patterns, Fraud patterns change dynamically and need to handle huge data. The purpose of this project is to develop a prototype or proof-of-the-concept ATM fraud detection framework, which is going to be useful for the Indian banking industry.
A framework titled “ATM Fraud Detection framework - ATMFD” is being developed, which is a Zero-Code Automated Machine Learning Framework that works with tabular transactions data for data analytics and predictive modeling. It abstracts the common way to exploratory data analysis, sampling for handling data imbalance, machine learning-based fraud detection model building, automatic hyper-parameter optimization, stacking of multiple models, near-real-time inference engine, and a dashboard for visual insights of the predictions. It is a no black-box framework as a user can choose any algorithm or method at every stage of data analytics and machine learning life cycle.
- Data Connectors for CSV, JSON, MySQL and MongoDB
- Automatic exploratory data analysis and report generation
- Undersampling and oversampling
- Extensive set of ML/DL algorithms
- Automatic Hyperparameter optimization
- Batch & Near-Real-Time Inference Engine
- Dashboard for visual insights of inferences
- UI for Multi-user and Multi-Job management
- Experiments tracking
Chief Investigator Details
Mr. Ramesh Naidu Laveti