DR.MAHESH,GPVS. KARTHIK,G. KISHORE,B. ABHINAV BALAJI, SUMITH RATHOD 03 03 The increasing volume and speed of network traffic along with the growing sophistication of cyber threats have made realtime intrusion detection essential for modern digital infrastructures Traditional security systems often depend on static rules or offline analysis which limits their ability to detect new or rapidly evolving cyberattacks This paper proposes a hybrid architecture that combines stream processing frameworks with deep learning models to enable realtime threat detection from network logs The system uses Apache Kafka for efficient log ingestion and Apache Flink for realtime stream processing and analytics Deep learning models including Long ShortTerm Memory LSTM networks and onedimensional Convolutional Neural Networks 1D CNN are applied for anomaly detection and threat identification The proposed approach is evaluated using benchmark datasets such as CICIDS 2017 and UNSWNB15 Experimental results show that the system can detect network threats with high accuracy and low latency while maintaining scalability under highthroughput conditions The architecture is therefore suitable for deployment in realworld operational environments where fast and accurate threat detection is crucial Copyright (c) Journal of Science Engineering Technology and Management Science. All rights reserved 2026