Data Connector predictive safety analytics and benchmarking
Learn how Geotab predicts collision risk and ranks fleet safety.
This article explains two parts of fleet safety in the Geotab Data Connector: predictive safety analytics and safety benchmarking. Predictive safety analytics uses driver behavior data and machine learning to anticipate and help prevent safety incidents. Safety benchmarking compares your fleet's safety performance against similar fleets.
If you are already a Geotab Data Connector user, the safety tables are available through the Data Connector. If you are building your own dashboard, you can build dashboard components based on the new data, similar to the Vehicle KPI tables. If you are using the templates, download the new template, and it populates after you make a successful connection.
The predictive safety analytics model focuses on the possibility of a collision, rather than a score. The Driver Safety Scorecard score weights are configurable and can be modified. With the predictive model, the score weights cannot be modified, because each subcategory has its own risk detail. The predictive model includes data from January 2, 2023 onward.
- Predictive safety analytics - Learn how Geotab calculates a fleet's predicted collision rate.
- Safety benchmarking - Learn how Geotab compares your fleet to similar fleets.
- Predictive safety tables - Review the FleetSafety_Daily, VehicleSafety_Daily, and DriverSafety_Daily table columns.
- Safety Benchmark template - Review the ready-made dashboard template for this data.
Device eligibility
All devices are eligible for these insights. The following device exclusions apply:
- Devices that have not communicated in the past month, or have not had at least 50 km of activity in the last 7 days, but still have a predicted collision rate.
- Devices that did not drive or record valid GPS readings.
- OEM, third-party, or government devices.
- Databases in the US Commercial Government instance.
- Devices with Restricted Data Mode activated.
Predictive safety analytics
Learn how Geotab predicts the risk of a collision for your fleet.
Predictive safety analytics uses driver behavior data to anticipate and help prevent safety incidents, instead of only reacting to them after they happen. Geotab monitors driving events such as hard braking, hard acceleration, sharp cornering, and speeding to calculate a fleet's predicted collision rate.
How Geotab calculates predictive safety analytics
For each vehicle, Geotab calculates a rate for each driving event and compares it to the total distance traveled.
Geotab groups vehicles with similar driving behavior to understand the average collision rate for each group. Vehicles are divided into groups based on their performance in hard braking, hard acceleration, cornering, and speeding. For example, Geotab creates ten groups for hard braking: the first group contains the 10% of vehicles with the fewest hard braking events, and the remaining vehicles are grouped the same way.
Geotab analyzes the average collision rate within each group to build a formula that predicts the likelihood of a vehicle being involved in a collision. The formula considers the rate of each exception event and the data from the vehicle's group.
Geotab combines these vehicle-level predictions into a fleet-level prediction using weighted averages. For example, a fleet's predicted collision rate might be 1.2 collisions per million miles, based on vehicle performance over the past week.
GPS and accelerometers are two common data sources used to infer harsh driving events. Accelerometers are effective at capturing abrupt movements, such as collisions and road impediments, but these sudden acceleration changes do not always indicate unsafe driving behavior. GPS data is more stable, because of its frequency and because Geotab\u0027s GPS speeds are validated against the vehicle\u0027s engine road speed. Geotab uses accelerometer data and statistical distributions alongside GPS data to determine harsh events and their thresholds, combining both data streams in a data-based approach.
Safety scores are updated daily at 23:30 UTC.
If a device is swapped to a different vehicle, no predictions are provided for either vehicle that day.
If a GO device is swapped from one MyGeotab database to another, the data is set to null on the day the swap occurs. The next day, only the new device ID records safety ranks, collision probabilities, and predicted collisions.
Safety benchmarking
Learn how Geotab ranks your fleet against similar fleets.
Safety benchmarking compares your fleet's safety performance to similar fleets in the system, so you can see how your fleet's risk level compares to the industry standard.
Comparable fleets share characteristics such as vehicle type, vocation, make, and model. Geotab groups these fleets into peer groups using a clustering algorithm.
Benefits of safety benchmarking
- Comparison with relevant peers.
- Goal-setting and progress monitoring.
- Comparative safety analysis of areas that need improvement.
Fleet clustering groups fleets with a similar composition of vehicle types and vocations together. Your fleet is benchmarked against other fleets with a similar mix and variety of vehicle types and vocations.
How Geotab calculates peer group benchmarks
Geotab compares each vehicle to about 1,000 similar vehicles, known as its peer group. A percentile rank compares a vehicle's predictions to the members of its peer group. For example, a rank of 0.01 means a vehicle performs better than 1% of its peer group, while a rank of 0.99 means it performs better than 99% of its peer group.
A vehicle's benchmark is the average predicted collision rate within its peer group. The peer group leader is the 20th percentile of the predicted collision rate within the peer group, meaning the leader has a lower predicted collision rate than 80% of the peer group.
For example, on June 21, 2023, a vehicle's predicted collision rate was 1.41 collisions per million kilometers, and its peer group benchmark was 1.47 collisions per million kilometers. The vehicle performed slightly better than the peer group average, but worse than the peer group leader. Its percentile rank was 0.54, slightly above the midpoint of 0.5, but considerably lower than the leader's rank of 0.8.
The number of fleets or vehicles in a cluster can change from day to day. One reason is that the number of active vehicles changes daily. For example, there are typically fewer active fleets or vehicles on weekends than on weekdays. Percentile ranks are not affected by the size of a cluster, so your rank always reflects your relative position within your peer group. The other reason is that Geotab retrains its similarity model weekly, so the composition of fleet clusters changes slightly from week to week to keep you compared against the fleets or vehicles most similar to you in location, vehicle type, and vocation.