Machine Learning & Anomaly Detection on IoT Traffic

Security framework for IoT

As far as an IoT infrastructure is growing, improving the the quality of service and monitoring potential attacks is crucial. However, when we are facing millions of payload per days, it becomes complicated for human to be aware of each micro change who can prefigure an incoming denial.

With Philéas, we’re convinced that anticipation is the next key on monitoring IoT traffic before it’s too late. We believe that using the latest technologies, thinking differently combined with our expertise is the best way to reach your goals.

Anomaly Detection on IoT Traffic

The challenge when you have an existing IoT infrastructure with an ever growing number of devices to manage it’s important for you to provide an excellent quality of service and being able to anticipate network malfunctions is now becoming a key element. After all keeping your customer happy is the aim of your services. However, It is very difficult / nearly impossible for you to put a filter and check every single use case of attack and/or malfunction.

Anomaly Detection Graph n°1 Anomaly Detection Graph n°2

IoT monitoring tools & platform

As we focus on anomaly detection, the data visualization through dashboard is crucial to keep you notified and have a clear view on what’s going on on your IoT traffic. Despite spending significant time monitoring analytics dashboards, we often missed technical incidents, but also business opportunities such as not knowing that a feature really moved the needle or that there was sudden adoption for a specific user group. The dashboard solutions can be implemented to your current infrastructure, or we can propose Jupyter notebook and datadog as default dashboards.

Anomaly Detection Graph n°3 Anomaly Detection Graph n°4

Machine learning - IoT Protocole & Real-Time monitoring

To fulfill these specific requirements and meet at the same time your business needs, our approach with Phileas is based on machine learning in order to train a neuronal network and record his natural traffic patterns. When something unusual/unexpected happens, we can report the issue quickly and tell you where you would need to intervene. 90% of the work is to understand the data, metadata and the enterprise logic. this is the First step.

Anomaly Detection Graph n°3 Anomaly Detection Graph n°4

Machine Learning on IoT traffic

Many machine learning models already are on the market, however, in order to get a fine tuned solution an complex mathematical understanding is needed in order to create an effective custom IoT Traffic Model. The second step is to train your neuronal network with two datasets. One dataset to feed the evolutive, learning algorithm and another dataset in order to test the effectiveness of your model.

When these steps have given you a result that you are happy with we pass on to the next and final step the third step consist in building the infrastructure so you can follow up your real time datastream by monitoring, alerts and reportings. To do so, Phileas give you access to last academic research on IoT behaviours analytics with Machine Learning. Severals mathematical models are adapted and tuned for IoT behaviours, forming a framework of good practice knowledge.

Data Exploration with Phileas