Title: Artificial Intelligence-Driven Cybersecurity: An Adaptive Framework For Network Anomaly And Ddos Detection
Authors: Dr Neha Paliwal
Volume: 10
Issue: 9
Pages: 132-141
Publication Date: 2026/09/28
Abstract:
Cybersecurity environments are becoming increasingly dynamic as cloud services, Internet of Things (IoT) devices, mobile platforms and distributed applications generate large and heterogeneous volumes of network activity. Traditional signature-oriented security controls remain valuable for known threats, but they can be less effective when malicious behaviour changes rapidly or resembles legitimate traffic. This study develops an expanded analytical framework for Artificial Intelligence (AI)-enabled cybersecurity, with particular attention to machine learning, deep learning, autoencoders, anomaly detection and Distributed Denial-of-Service (DDoS) detection. The work is grounded in the supplied research article and supplements its analysis with current external evidence from ENISA and NIST and with verified descriptions of widely used intrusion-detection datasets. The original source reports 15 comparative accuracy values ranging from 82.3% to 99.95%; secondary statistical analysis of those values gives a mean of 96.42%, a median of 98.10% and a range of 17.65 percentage points. These figures are treated as reported results, not as newly collected experiments. The study further proposes a human-in-the-loop architecture in which network traffic is captured, cleaned, normalized and represented through an autoencoder before anomaly scoring, classification and analyst verification. Current ENISA evidence indicates that DDoS represented 77% of reported incidents in its 2025 threat landscape covering 4,875 incidents from July 2024 to June 2025. The analysis demonstrates the potential of AI while emphasizing that accuracy alone is insufficient for cybersecurity evaluation. Reliability, false-positive control, explainability, computational cost, concept drift, adversarial robustness and human oversight must also be considered. The article concludes with practical recommendations for multi-dataset validation, continuous monitoring, explainable AI and responsible integration with established cybersecurity risk-management frameworks.