Dr Nebojsa Bacanin
15. окт 2025.
Achieving 97.96% accuracy in threat detection through a hybrid CNN-Boosting architecture and Explainable AI.
The backbone of any Smart City is the Internet of Things (IoT). These interconnected devices optimize everything from traffic flow to energy usage. However, there is a catch: most IoT devices have "lightweight" hardware, meaning they lack the processing power for heavy-duty built-in security.
Because these devices are constantly online and often under-protected, they have become prime targets for cyberattacks. Standard firewalls are no longer enough; we need dynamic, AI-driven countermeasures.
A Two-Stage Shield: The Composite Architecture
This research introduces a sophisticated, two-tier defense system designed to catch intruders before they can compromise urban infrastructure.
Tier One (The CNN Layer): The first stage uses an optimized Convolutional Neural Network (CNN). While CNNs are famous for image recognition, here they are used to identify complex patterns within network data traffic.
Tier Two (The Boosting Layer): The system then passes data to advanced machine learning classifiers—XGBoost and AdaBoost. These are further refined using metaheuristic optimization to ensure they are picking up on the most relevant "features" of a cyberattack.
Record-Breaking Accuracy
Using a real-world IoT dataset, the framework was put to the test against multi-class attacks (various types of hacking attempts). The results were industry-leading:
Performance: The top configurations achieved a 97.96% accuracy rate in identifying specific types of intrusions.
Efficiency: By using feature selection in the second tier, the model ignores "noise" and focuses only on the data points that indicate a real threat.
Opening the "Black Box" with XAI
One of the biggest hurdles in medical or municipal AI is trust. If an AI blocks a system, engineers need to know why.
This study utilizes Explainable Artificial Intelligence (XAI) to interpret the model’s reasoning. Instead of being a "black box" that simply says "Yes" or "No," the system provides insights into its decision-making process. This allows cybersecurity teams to refine their protocols and better understand how threats are evolving.
Conclusion
As our cities become "smarter," our security must keep pace. By merging deep learning with optimized boosting classifiers, this research provides a blueprint for a secure, resilient operational framework for the future of urban living.