Anomaly Detection in Enterprise Networks Using Deep Neural Networks and Real-Time Streaming Data

Abstract

Modern enterprise networks generate massive volumes of traffic data, making traditional rule-based intrusion detection systems increasingly inadequate against sophisticated and evolving cyber threats. This paper presents a deep neural network (DNN)-based framework for anomaly detection in enterprise networks, leveraging real-time streaming data to identify malicious activities and network intrusions with high precision and minimal latency. The proposed system integrates Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs) to capture both temporal dependencies and spatial patterns within network traffic flows. By employing Apache Kafka and Apache Flink as the real-time data streaming backbone, the framework ensures scalable, fault-tolerant ingestion and processing of high-velocity network telemetry. Feature engineering techniques, including statistical flow analysis and payload inspection, are applied to enrich input representations fed into the model. The system is trained and evaluated on benchmark datasets, namely NSL-KDD and CICIDS2017, demonstrating superior detection accuracy, reduced false positive rates, and improved adaptability to zero-day attacks compared to conventional machine learning approaches. Experimental results confirm that the proposed architecture achieves over 98% detection accuracy while maintaining real-time processing throughput suitable for enterprise-scale deployments. This work establishes a robust, intelligent, and scalable solution for proactive cyber threat detection in dynamic network environments.