Smart cameras, doorbells, sensors and plugs are tiny computers on the internet, and most are too weak to defend themselves. Deep-IDS is an AI guard that sits between these devices and the internet, spotting attacks in about a second and a half.
The problem
The Internet of Things is everywhere, from homes to factories. But IoT devices have little memory or processing power, so they cannot run advanced security software. Attacks against them keep rising.
Much research builds ever more complex detection systems, but many are too heavy to run in the real world.
What the research did
We designed Deep-IDS to be practical first. It uses a compact LSTM network, a type of AI that learns patterns in sequences, with 64 units, trained on the widely used CIC-IDS2017 network attack dataset.
Deep-IDS runs on an edge server: a small computer close to the devices, acting as a gatekeeper between them and the internet. It watches for five common attacks: denial of service (DoS), distributed denial of service (DDoS), brute-force password guessing, man-in-the-middle and replay attacks. We also carefully balanced the detection rate against the false alarm rate so it works in real time.
What we found
Deep-IDS detected 96.8% of attacks with an overall accuracy of 97.67%, precision of 97.67%, recall of 98.17% and F1-score of 97.91%. On average it needed 1.49 seconds to identify and stop an intrusion, blocking the source of the malicious traffic.
By protecting the devices, it also protects the networks they are connected to.
Who benefits
- Households with smart devices
- Businesses and factories running networks of sensors
- Network providers looking for lightweight protection at the edge
Limits and what comes next
Deep-IDS was trained and tested on a benchmark dataset from 2017. New attack types appear constantly, so it would need retraining on fresh data, and real-world trials, before wide deployment.
The team
This paper was led by Sandeepkumar Racherla and Prathyusha Sripathi, data scientists at Amazon, with me, Md Alamgir Kabir (Mälardalen University, Sweden), Md Whaiduzzaman (QUT) and Syed Aziz Shah (Coventry University, UK). With 127 citations, it is one of my most-cited papers.
Frequently asked questions
Can smart home devices be hacked?
Yes. Many have weak security because they have limited computing power. A network-level guard like Deep-IDS can watch traffic on their behalf.
What attacks does Deep-IDS detect?
DoS, DDoS, brute-force, man-in-the-middle and replay attacks.
How fast is Deep-IDS?
It identifies and blocks an intrusion in 1.49 seconds on average.