Real-Time Fraud Detection Neural Network Pipeline

Construct a PHP-based neural network pipeline for real-time financial fraud detection that can process multiple data streams simultaneously. Implement a machine learning model that can adapt to emerging fraud patterns, with support for transfer learning and anomaly detection across payment networks. The system must achieve 99.5% accuracy, support multiple data ingestion methods, and provide a flexible feature engineering framework that can be quickly updated by data scientists.