How does machine learning contribute to WirelessCar’s development of connected car services?

June 29th, 2023

Machine learning plays an essential role in the development of connected cars and the digital services that allow these software-defined vehicles to reach their full potential. In this article, we take a closer look at WirelessCar’s work with machine learning, how that work contributes to new or improved connected car services, and the role of the OEMs in this process.

What is machine learning?

Machine learning is a type of artificial intelligence where computer algorithms are trained to learn from sets of data, instead of being hard coded (thus finding the right strategy from the training data). We use the datasets to train machine learning models.

In “traditional programming”, a programmer will break down an issue or problem, find the underlying patterns, and instruct a program on how to solve the issue. With machine learning, we provide the model with all the relevant input, as well as the correct results we want it to predict. The model will then work out the pattern, what is causing the issue/problem, find a way for accurately predicting the results and use that to generate answers.

How can machine learning improve connected car services?

In the field of connected cars and connected car services, machine learning is particularly useful for data analysis. Connected cars and their sensors provide enormous amounts of data; data which can be analyzed and used to improve virtually every aspect of car ownership, fleet management, and mobility at large. Some examples:

Anomaly detection – machine learning can help identify issues by detecting unusual behavior. By training on both normal and failure patterns using DTC (Diagnostic Trouble Code) reports, it is able to identify deviations, which can result in quicker diagnoses, improved vehicle reliability, and reduced maintenance costs.

Better digital services for drivers – drivers’ preferences and user patterns (while not collected or analyzed in individual detail) can contribute to making connected car services better and more accessible; through optimized routing or improved EV charging, for example.

Safety and security – machine learning helps detect risks to driver safety and automotive cybersecurity. Maintenance needs can also be predicted and met more efficiently.

rear view of a car with abstract data visualization

WirelessCar’s data science and machine learning processes

So, what does the greater data science process look like in our case, and where does machine learning come into the picture?

It all starts with the identification of a problem – an issue that needs to be solved. Depending on what the problem is, and the connected car data at our disposal, we determine which machine learning technique we should use (more on that below).

Once collected, the connected car data is prepared, analyzed, and processed to fit the machine learning technique we want to use. The data and the machine learning model must be relevant to the task at hand, so that they can provide actual answers to the problems we seek to solve.

The machine learning models are then put to work, a process known as training. The models receive the prepared data, find patterns, make predictions, and learn from it – getting better as they go along. Once we have a result, we analyze it and discuss with the stakeholders involved on how best to apply it in our work.

Which machine learning techniques does WirelessCar use?

WirelessCar currently employs all the three main types of machine learning: supervised learning, unsupervised learning, and reinforcement learning.

Supervised learning is used when we are trying to train a model to solve an issue using labeled data (thus data tagged with known, correct information). This is a good method to use as long as the data is of good quality and relevant to the problem we are trying to solve. A concrete example: Engine failure detection, based on diagnostic trouble codes (DTC) with known failure and reparation dates.

Unsupervised learning, by contrast, is used when we need to come to a conclusion by working with unlabeled data. The machine learning model is required to analyze and find patterns from this data, such as when we try to categorize drivers into different groups according to certain parameters. We may not initially know how many groups there will be, nor how best to divide them up. Through unsupervised learning, we can find and evaluate the type of clustering/categorization that makes the most sense for our specific use-case.

Then there is reinforcement learning, which means that the machine learning model learns through continuous feedback on what it does. Take autonomous driving: the model can be given rewards or penalties depending on how well it performs/”drives”. If it drives into an obstacle, it will get minus points and thus learn to avoid doing that in the future, for example.

group of engineers working on 3D car

Ensuring the security and reliability of our machine learning work

WirelessCar has chosen to work with Amazon Web Service (AWS) as our data science and machine learning partner. Our partnership with AWS goes back a long time, and our data is stored in AWS’s secure cloud services. We use Amazon SageMaker and Redshift for our machine learning models and QuickSight for visualization.

It is crucial to stress that the data we analyze is anonymized. We do not have any traceable user information (drivers’ social security numbers, vehicle numbers, etcetera) and thus cannot use it in our machine learning. Additionally, our machine learning work does not involve that kind of granular breakdown of data. We are interested in the bigger picture, and in creating products and solutions that are useful for all OEMs, in all their markets.

Overcoming the challenges of machine learning, together with OEMs

What are the main challenges of machine learning, at least in this context? How can we overcome these challenges, and deliver better connected car services for OEMs?

1. Connected car data collection and analysis
Essentially, data quantity is the key to connected car service quality. The more connected car data you can work with, the better analyses you will be able to make through your machine learning processes. Not all data will be relevant for a particular machine learning project, but as connected car services inevitably become more interconnected and complex, it is critical to have access to as much data as possible when developing these services.

2. Handling big data volumes
Even comparatively small data sets can concern millions of vehicles, so handling enormous amounts of data is highly complex. At WirelessCar, we make sure to only invest in and work with the kind of technology that can live up to these demands, and that provides us with the tools necessary for big data management. That is a prerequisite for continuously enhancing our machine learning, and developing our connected car services for (and alongside) OEMs.

3. Collaborating closely with OEMs
The closer the collaboration between us and the OEM, the better the results. Rarely is this more true than in the case of machine learning. We need to understand our customers’ needs and problems (and their customers’, in turn), and train our machine learning models accordingly. Through this collaborative process, not only are we able to overcome multifaceted issues together, but better prepare for the future as well.

If you want to know more about how our work with machine learning can help your business offer, do not hesitate to reach out to me: you will find my contact information below. Machine learning is a vast and fascinating subject, one we will return to in upcoming articles. In the meantime, make sure to check out other, related articles here on our WirelessCar Insights blog – on mobility insights and sustainability, EV driver insights, the FREEDOM project, automotive cybersecurity and data privacy, and lots of other topics!

Sami Fatmi
Data Scientist