Deep Neural Network and Reinforcement Learning Algorithms for Real-Time Threat Detection and Autonomous Cyber Defense in AI-Driven Internet Services – Journal of Internet Services and Information Security

Volume 16 - Issue 1

Deep Neural Network and Reinforcement Learning Algorithms for Real-Time Threat Detection and Autonomous Cyber Defense in AI-Driven Internet Services

Vijaylaxmi Utkal Patil Assistant Professor, Department of Information Technology, Kasegaon Education Society's, Rajarambapu Institute of Technology, Affiliated to Shivaji University, Sakharale, India.
vijayalaxmi.patil@ritindia.edu 0009-0008-7755-0672
Dr.K. Hussain Associate Professor, Head, Department of Electrical Engineering, Sharad Institute of Technology, College of Engineering, Ichalkaranji, Kolhapur, Maharashtra, India.
hussain16679@gmail.com 0000-0001-9744-4764
Raghavendra Reddy Assistant Professor, Department of Electrical and Electronics Engineering, Sapthagiri NPS University, Bangalore, Karnataka, India.
raghavendrareddy@snpsu.edu.in 0009-0003-7139-6620
Dr.H. Krishnamurthy Associate Professor, Department of Computer Science Engineering, Sapthagiri NPS University, Bangalore, Karnataka, India.
krishnamurthy.h8@gmail.com 0000-0002-3824-484X
Dr.J. Narendra Babu Professor, Department of Computer Science Engineering-Data Science, Sapthagiri NPS University, Bangalore, Karnataka, India.
drjnbabucse@gmail.com 0009-0002-1235-620X
H.B. Ashwini Gowda Assistant Professor, Department of Electronics and Communication Engineering, Dayananda Sagar College of Engineering, Bengaluru, Karnataka, India.
ashwini.hb7@gmail.com 0009-0000-5240-4179

DOI: 10.58346/JISIS.2026.I1.042

Keywords: Deep Neural Networks, Reinforcement Learning, CNN–LSTM, Deep Q-Network, Real-Time Threat Detection, Autonomous Cyber Defense, AI-Driven Internet Services.

Abstract

The swift growth of Internet services based on AI has amplified the threat to cybersecurity to a great extent, as it requires intelligent and autonomous defense systems rather than conventional rule-based systems. This paper suggests a combined Deep Neural Network-Reinforcement Learning (DNN-RL) architecture to control the detection of threats in real-time and adjust the cyber defense. The architecture integrates a hybrid CNNLSTM model for spatial-temporal traffic analysis and a Deep Q-Network (DQN) agent for optimal mitigation decision-making. The UNSW-NB15 dataset, as well as the CICIDS2017 dataset, consisting of more than 5.3 million labelled network flow records, were experimented on. The offered framework had a higher ROC-AUC of 98.1%, a false alarm rate of 3.8%, a defense success rate of 92.4%, and a network throughput of 742 Mbps than traditional frameworks like Random Forest and Support Vector Machines on all metrics. The reinforcement learning module minimised the mitigation latency by about 22.5% relative to the detection-only CNNLSTM model and enhanced the adaptive response efficacy during multi-stage attack situations by 7.7%. Paired t-tests were used to statistically validate the results that the performance had improved significantly at p < 0.01. The findings reveal that the combination of deep learning-based detection and reinforcement learning-based autonomous response can increase operational efficiency, resilience, and scalability under dynamic AI-enabled service environments. The framework facilitates the ongoing learning process and dynamic mitigation, and it is appropriate to be implemented in large-scale, real-time cyber defense infrastructures.

Date

February 2026

Page Number

734-746