UAE engineer develops AI that detects cyberattacks 31 times faster

Abu Dhabi: An Emirati engineer has developed an artificial intelligence (AI) system that detects cyberattacks with 97.45 per cent accuracy, identifying 14 types of threats in just 0.87 milliseconds — around 31 times faster than a previous international research benchmark tested on the same device.

Mohammed Al Shehhi, a data scientist and lecturer at the Higher Colleges of Technology, developed the system to help protect internet-connected devices, including home surveillance cameras, smart appliances, sensors and industrial equipment. By reducing the computing power and memory required for threat detection, the technology could pave the way for more efficient cybersecurity solutions for devices with limited processing capabilities.

Al Shehhi will present his findings today at the 12th World Internet of Things (IoT) Congress, taking place at Khalifa University in Abu Dhabi from October 26 to 29. The UAE is hosting the international conference for the first time, bringing together researchers and experts from academia, industry and government to discuss artificial intelligence, cybersecurity and smart cities.

In an exclusive statement to Gulf News Al Shehhi explained that the research uses explainable artificial intelligence to identify the network indicators most useful for detecting malicious activity, rather than analysing every available feature.

This approach reduced the number of features used by one model from 47 to 20 a reduction of approximately 57 per cent- while simplifying the detection process.

The main model was also reduced to approximately 301 kilobytes, compared with 5,213 kilobytes for the full model used as the study’s starting point.

Researchers developed a smaller version measuring about 152 kilobytes, with a prediction time of approximately 0.61 milliseconds, while maintaining strong performance in testing.

These reductions could make AI-based security more practical for connected devices with limited memory and processing power.

The research was evaluated using established benchmark datasets commonly used in internet of things security studies. Al Shehhi also tested the model on an actual small computing device to assess its performance under practical hardware constraints.

A multi-round evaluation found that the model outperformed the strongest published study with comparable constraints included in the comparison on one performance metric, scoring 0.969 against 0.947 — a difference of 2.2 percentage points.

The findings demonstrate the methodology’s potential across detection accuracy, prediction speed and model size.

However, the technology remains at the research and development stage and has not yet been commercially deployed in homes or industrial facilities. The results are based on scientific tests using benchmark datasets and physical hardware, and further development is required before commercial use.

As the number of connected devices grows, so does the need for cybersecurity solutions that can operate within tight hardware limits. Devices such as surveillance cameras, smart refrigerators, washing machines, sensors and industrial equipment often have limited memory and processing power to run conventional security software.

Al Shehhi’s research could help pave the way for lightweight intrusion detection systems for smart homes, factories, industrial facilities and smart cities, enabling devices to identify suspicious network activity while using fewer computing resources.

Al Shehhi’s presentation at the 12th World Internet of Things Congress showcased UAE-developed research to an international audience, highlighting the contribution of Emirati researchers to advances in artificial intelligence and cybersecurity.

The study also underscores the role of UAE higher education institutions in developing research talent and translating academic knowledge into practical solutions. By addressing real-world cybersecurity challenges, the research demonstrates how Emirati expertise can contribute to developing future technologies.