Zurück zur
Übersicht
20
.
11
.
2020

Transport-mode recognition and the driver/passenger dilemma

How Swiss Re helps insurers understand their clients' mobility habits with UBI propositions
Benedetta Cerruti
Scoring & Analytics Expert at Swiss Re
Elia Giacomini
Data Scientist at Swiss Re

Customer mobility is changing. How can insurers adjust?

The way people travel is undergoing fundamental change. New generations of commuters are increasingly moving away from owning a car, instead switching between modes of transport best suited to their travel needs and purposes.
It's time for insurance to adapt to this more flexible and fluid mobility space and make the shift from static, traditional risk assessment to tailor-made profiling based on each individual's unique characteristics and driving style. It's time for insurance to embrace customer-centric, dynamic, usage-based insurance (UBI) models.

Figure 1: The most-used modes of transport, according to Coloride's global data from 2019, with differentiation between car drivers and passengers.

Swiss Re's telematics app Coloride helps recognise the mode of transport

Innovative, customer-centric UBI insurance models often involve the use of smartphones and their sensors. One of the challenges of smartphone-powered motor insurance offerings is recognising whether users are travelling by car, bus, train or bicycle, or combining various modes of transport in a single journey.
That's where Coloride, the telematics app developed by Swiss Re and its tech subsidiary Movingdots, comes in. There's no need to install additional hardware in the vehicle. Leveraging smartphone sensors, data is collected and anonymised for the best privacy protection to enable driving-style assessment and support various UBI offerings (such as Pay-How-You-Drive). Thanks to the latest machine-learning models, Coloride can transform raw data from the sensors into powerful insights. In an urban environment, the speed of travel of a vehicle is not always the determining factor: think about slow-moving rush-hour traffic. The most discriminative features come from the smartphone's three-axis accelerometer and the built-in GPS. In order to boost the predicting power of the transport-mode recognition algorithm, the data is enriched with trip-contextual information from our map data provider HERE, as explained in more detail in this article. All this information is then fed into a state-of-the-art implementation of gradient-boosting decision trees, chosen for their ability to grasp data correlation structures. This lets Coloride detect the main modes of transport with high levels of accuracy.

Table 1: Coloride detection accuracy for different modes of transport.

Models learn and improve with every trip

The reliability of the automated transport-mode recognition increases as more data points are accumulated. If the model fails to recognise the mode of transport correctly, users have the option to manually correct the predicted transport mode in Coloride. The changes are automatically detected and the supervised model is retrained in order to avoid repetition of the same mistake and improve the overall performance. The model's prediction capabilities improve in a continuous cycle.

Figure 2: Machine-learning model based on continuous improvement through new available data.

The key to UBI product design: understanding the "driver fingerprint"

One of the challenges in designing UBI programs is understanding which of the detected and recorded car trips are being taken by the insured driver or car. If the program detects that the user is travelling by car, then we want to know the user's role. Are they the driver, or are they just sitting back and enjoying the ride as a passenger in a taxi or a friend's car? Coloride's driver/passenger detection models help answer this question.
Analysing users' recurring activities plays a significant role in creating a successful model: taking the same bus to the office every day or following the same route to the gym every week are important routines. Coloride uses advanced unsupervised clustering techniques to group recurrent trips, so whenever similar trips are recorded, the app can assess the mode of transport with a very high probability. But that's not all. We make full use of smartphones' high-tech sensors, which can record physical quantities like acceleration at high frequency (as high as 100 Hz), to extract various features that help to discriminate between different driver/passenger situations. Coloride does this by applying a combination of physics and machine learning to a big data collection of hundreds of thousands of trips, allowing it to correctly distinguish between a driver and a passenger in nearly 85% of cases.

A new technological approach to connected mobility

Understanding mobility patterns will help insurers to better serve their customers and to design appropriate cover for them. With Coloride featuring automated transport-mode detection and driver vs passenger recognition among its capabilities, insurers can fully count on Swiss Re to get reliable input and determine the right price for the coverage. Insurers can also use these insights to unlock the various benefits of a UBI solution, such as driver coaching or distracted-driving detection (a more comprehensive overview can be found in this publication). And while the journey usually starts with creating a UBI product, insurers can make the most of Swiss Re's expertise to enable more comprehensive connected mobility solutions.

Swiss Re: helping insurers improve mobility safety as a leading partner in P&C product innovation

As a leading partner in the P&C product-innovation space for insurers across the globe, Swiss Re has developed advanced machine-learning and user-profiling techniques within its telematics proposition with the goal of supporting insurers in their journey towards safer mobility. Coloride is available in nine languages, and is already used by many insurance companies across the world. Have we sparked your interest? Get in touch with us!

Artikel teilen