UAE engineer develops AI that detects home camera cyberattacks 31 times faster than global benchmarks

Abu Dhabi: Emirati engineer Mohammed Al Shehhi, a data scientist and lecturer at the Higher Colleges of Technology, has successfully developed an innovative technology based on artificial intelligence to detect cyberattacks that may target surveillance cameras inside homes and smart devices connected to the internet. His work has achieved an accuracy of 97.45% and a speed in detecting threats that is about 31 times faster than previous international research reference that used the same device, in a scientific achievement that reflects the ability of Emirati minds to compete with global research in the fields of artificial intelligence (AI) and cybersecurity. In a special statement, Al-Shehhi revealed details of his invention, which allows for the detection of cyberattacks at extremely high speeds, without the need for large operational capabilities. This paves the way for the development of future solutions that enhance the protection of smart devices, which have become part of the daily lives of families, and confirms the ability of national talents to move from using technology to contributing to its development and creating solutions. Al Shehhi will present the results of his research during the 12th World Forum on Internet of Things (WF-IoT), which Abu Dhabi is hosting for the first time in the UAE, from October 26 to 29, at Khalifa University, with the participation of researchers and experts from academic institutions, industrial and governmental sectors, to discuss the latest developments in artificial intelligence, cybersecurity and smart cities. The achievement is of particular importance as it is the product of scientific research developed by a national expert within one of the UAE’s higher education institutions, and has achieved advanced results in a field that is witnessing rapid international competition, reflecting the ability of national researchers to move from using modern technology to participating in its development and producing scientific knowledge. Al Shehhi explained that the expansion in the use of internet-connected devices, from home surveillance cameras, smart refrigerators and washing machines to sensors and industrial equipment, poses increasing challenges related to protecting these devices from cyberattacks, especially since many of them do not have sufficient memory or processing power to run large-scale traditional protection systems. He pointed out that the challenge addressed by the research is the development of an artificial intelligence-based system that can distinguish between normal network traffic and cyberattacks quickly and accurately, without the need for large operational resources, thus opening the way for the development of future solutions to protect smart devices with limited capabilities. He explained that the research methodology he developed is based on interpretable artificial intelligence, and is based on identifying the most important information in detecting cyberattacks, instead of analysing dozens of indicators related to network traffic, which helps to reduce the size of the model and speed up its performance, while maintaining a high level of accuracy. The methodology contributed to reducing the number of features that one of the models relies on from 47 to only 20 features, a decrease of approximately 57%, without the need to use all the indicators that the original model deals with, which enhances the possibility of developing systems for detecting cyberattacks that are suitable for small devices. Al Shehhi revealed that the test results showed the main model’s ability to distinguish between normal network traffic and 14 types of cyberattacks, with an accuracy of 97.45%, while it recorded a prediction time of about 0.87 milliseconds on a small device, compared to about 27 milliseconds in a previous research reference that used the same device, making it about 31 times faster in terms of the measured prediction time. He explained that the results were not limited to the level of accuracy and speed, as the research also managed to reduce the size of the main model to about 301 kilobytes, compared to about 5,213 kilobytes for the full model that was used as a starting point in the study, while a smaller version was developed with a size of about 152 kilobytes, with a prediction time of about 0.61 milliseconds, while maintaining advanced research performance. To verify the methodology’s effectiveness, Al Shehhi stated that tests were conducted using standard data known in Internet of Things security research, along with running the model on an actual small computer, with the aim of measuring its ability to operate using limited operational resources, and not being satisfied with theoretical tests. He continued: The multi-round research evaluation showed that the model outperformed the strongest published study with limitations included in the comparison in one of the performance indicators, achieving a result of 0.969 compared to 0.947 for the