Open research Europe · 2026
SentinelSphere: An AI-driven cybersecurity platform integrating real-time threat detection with security awareness education.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Tantaroudas Nikolaos D, Karachalios Ilias, McCracken Andrew J
Аннотация
BACKGROUND: The cybersecurity domain faces dual challenges: a global shortage of qualified professionals and persistent human-factor vulnerabilities contributing to the majority of security breaches. Traditional Security Information and Event Management (SIEM) systems generate excessive false positives causing analyst alert fatigue, while conventional awareness programmes demonstrate limited effectiveness in changing user behaviour. Integrated solutions addressing both technical detection and human-centric education are needed.
METHODS: We developed SentinelSphere, an AI-driven platform combining machine learning-based threat detection with Large Language Model (LLM)-powered security training. The detection component employs an Enhanced Deep Neural Network trained on the CIC-IDS2017 and CIC-DDoS2019 benchmark datasets. To mitigate the substantial class imbalance present in these datasets (e.g. 168,186 benign samples versus 21 SQL injection samples), class weights inversely proportional to class frequencies were applied during training. The educational component utilises Microsoft's Phi-4 model (14 billion parameters) with quantisation techniques to enable deployment on standard hardware. System performance was optimised through Rust-based preprocessing and validated via pilot deployments across two workshops with a total of 79 stakeholders (3 professionals in maritime, educational and manufacturing sectors; 76 in an educational setting) in Greece.
RESULTS: The Enhanced DNN achieved high detection accuracy with significant false positive reduction compared to baseline models, while maintaining strong recall for critical attack categories including DDoS, botnet, and brute force attacks. Rust optimisation delivered substantial speedup in both single-record and batch processing. Stakeholder validation revealed high satisfaction rates, with most participants achieving improved security concept comprehension post-training. The platform identified critical awareness gaps, particularly regarding data protection regulations and multi-factor authentication adoption.
CONCLUSIONS: SentinelSphere demonstrates that integrating intelligent threat detection with adaptive, LLM-powered security education shows strong potential to address both technical detection and human security awareness challenges, although longitudinal studies are required to validate sustained behavioural change. The resource-efficient design enables deployment in SME environments without enterprise-grade infrastructure, supporting comprehensive cyber resilience approaches within the European Union's regulatory framework.
METHODS: We developed SentinelSphere, an AI-driven platform combining machine learning-based threat detection with Large Language Model (LLM)-powered security training. The detection component employs an Enhanced Deep Neural Network trained on the CIC-IDS2017 and CIC-DDoS2019 benchmark datasets. To mitigate the substantial class imbalance present in these datasets (e.g. 168,186 benign samples versus 21 SQL injection samples), class weights inversely proportional to class frequencies were applied during training. The educational component utilises Microsoft's Phi-4 model (14 billion parameters) with quantisation techniques to enable deployment on standard hardware. System performance was optimised through Rust-based preprocessing and validated via pilot deployments across two workshops with a total of 79 stakeholders (3 professionals in maritime, educational and manufacturing sectors; 76 in an educational setting) in Greece.
RESULTS: The Enhanced DNN achieved high detection accuracy with significant false positive reduction compared to baseline models, while maintaining strong recall for critical attack categories including DDoS, botnet, and brute force attacks. Rust optimisation delivered substantial speedup in both single-record and batch processing. Stakeholder validation revealed high satisfaction rates, with most participants achieving improved security concept comprehension post-training. The platform identified critical awareness gaps, particularly regarding data protection regulations and multi-factor authentication adoption.
CONCLUSIONS: SentinelSphere demonstrates that integrating intelligent threat detection with adaptive, LLM-powered security education shows strong potential to address both technical detection and human security awareness challenges, although longitudinal studies are required to validate sustained behavioural change. The resource-efficient design enables deployment in SME environments without enterprise-grade infrastructure, supporting comprehensive cyber resilience approaches within the European Union's regulatory framework.