nd-serge/mobile-money-fraud-detection

In many parts of Africa, mobile money services have become a primary solution for financial inclusion. However, fraudsters often exploit users who may have limited knowledge of cybercrime tactics. This project aims to develop a microservice that mobile money operators can use to identify and block fraudulent transactions in real time # mobile-money-fraud-detection ## Overview This project delivers a microservice for mobile money operators across Africa to identify and block fraudulent transactions in real-time. As mobile money services have become a primary solution for financial inclusion in many African regions, this system helps protect users who may have limited knowledge of cybersecurity from sophisticated fraud tactics. ## Objectives 1. Automate the training data pipeline 2. Deploy a microservice with API endpoints ## Data description This project utilizes the Synthetic Financial Datasets For Fraud Detection from Kaggle. The dataset simulates mobile money transactions based on a sample of real transactions extracted from financial logs from a mobile money service implemented in an African country. ## Training workflow This workflow automates the end-to-end process for mobile money fraud detection using three main steps: - `run_preprocessing` : Downloads the dataset from Kaggle and applies essential data transformations, including filtering and cleaning. This step prepares the data for model training. - `run_training`: Trains a machine learning model on the processed data. The training step includes hyperparameter tuning and evaluation on a test set. - `get_best_model` : Selects and registers the best model based on the highest average F1 score on the test data. The chosen model is then stored in the MLflow Model Registry for future use. Each step is orchestrated using Prefect and MLflow, ensuring reproducibility and traceability of experiments. Note: It is imperative to have a Kaggle account to download the dataset. ### Tools & Technologies - `Makefile`: Task automation and standardization - `Prefect`: Workflow orchestration - `MLflow`: Experiment tracking and model registry - `LocalStack S3`: Local S3-compatible storage - `SQLite`: Lightweight database - `Pipenv`: Python dependency management ### How to Run the Training Workflow *Make sure you are using Python 3.12.* 1. …