BlueGuard – Building Highly Accurate Classification Models

BlueGuard - Accurate Classification Models for Accelerated Decision-Making

Small Data Analytics provide accurate classification models that improve operational efficiency in any sector, from HR churn reduction to coastal security, and beyond.

We recently joined forces with Thales to detect real-world underwater threats. Our algorithm built errorless classification models powerful enough to treat noisy signals and suppress errors to ensure coastal security.

Not only does this use case demonstrate MyDataModels’ technological ability to respond to challenges around coastal security, but it can also be extended to a large variety of industrial use cases.

Improving coastal security

BlueGuard was Thales and MyDataModels’ answer to the Blue Innovation Challenge led by the city of Nice to improve coastal security. It combines Thales’s sonar technology, MyDataModels’ AI engine and the expertise and decision-making skills of coast guards.

BlueGuard aims to automate threat classification with trusted, frugal AI to accelerate detection and reduce decision-making times for improved coastal security.

In a high-risk context, ZGP creates real-time, accurate classification models to label potential underwater threats. Using sonar data with an accuracy of over 95%, BlueGuard provides coast guards with the best insights in under 30 seconds. They can then handle the underwater threat accordingly in under 10 minutes.

Coast guards confirm threats and launch the necessary interventions through User Interfaces based on real-time, accurate insights for augmented decision-making.

Accurate classification models with unbalanced data

Among various products, Thales manufactures land sonars that monitor coastal areas. They generate signals about potential underwater threats. Of course, most of the events these drones detect are not threatening. On the flipside of that, data sets are unbalanced while raw sonar signals are very noisy and poorly labeled. Furthermore, these time series files increase dramatically the volume of data.

Building a clean tabular Small Data set and training the models to detect any threat was the main challenge. Additionally, most of the potential threats were actually harmless, leaving little room for the algorithm to differentiate unbalanced outcomes. To improve coastal security, it was crucial to validate our models so that they don’t miss any false negatives, to detect all threats even if it means having more false positives. 

We implemented continuous training for models so that they improve with new incoming data. As time goes on, signals evolve and models need to constantly learn and improve to stay relevant.

Accelerate decision-making with Explainable AI

The results were quite spectacular. The predictive models we trained with this cleaned-up processed data showed recall and precision rates above 95%. In non-technical terms, the model returned no false negatives and reached coastal security objectives.

MyDataModels’ Explainable AI approach helped coast guards. They knew they could trust the AI’s accurate classification and get a clear understanding of any potential threat’s causes and degree of risk. It helped them anticipate security issues and mobilize the right resources accordingly. 

Getting instant information on the level and type of underwater threat is a game-changer in such a tense environment. In seconds, they understand what is dangerous and what is not. This was backed up by a real-life demonstration conducted alongside Thales in the area surrounding Nice airport in November 2021.

Extending Accurate classification to manufacturing

Of course, most organizations don’t face security challenges on this level. However, accurate classification models are highly valuable in other unbalanced data use cases, such as manufacturing. Take, for example, your production line. If it is well-designed, you probably don’t encounter significant problems very often, meaning you don’t have many samples to teach the algorithm to detect defective parts and predict when maintenance is required. When something does go wrong, though, it can cost you a lot of money!

Applying a similar accurate classification approach to these use cases will help you cut costs and improve your Return on Investment. It has proved to be valuable not only for the BlueGuard project but also for our manufacturing customers, so contact us now for a demo and a discussion about your projects!

Share

Start making sense of  your data

Test easily Quaartz with our test data here

You might also like...

MyDataModels among France Digitale’s top AI startups!

The CIAR Project: Frugal Data Analysis for Process Optimization

Frugal Data Analysis for Process Optimization

How Small Data Predicts Delivery Delays

How Small Data Predicts Delivery Delays

Improving Customer Satisfaction with Small Data Analytics

Improving Customer Satisfaction with Small Data Analytics