- Machine Learning
Across industries and supply chains, the safety of drivers and efficient use of fuel by truck fleets are an increasing concern. This project focused on understanding driving styles, understanding the tradeoffs between safe and efficient driving styles, and finding the highest levels of safety and fuel efficiency. We worked with Coca-Cola FEMSA to analyze one year of telematics data from over 3,000 vehicles. To analyze the data, we employed a methodology that involved multiple machine learning and analytical techniques, including multiple regressions, a random forest classification algorithm, Bayesian Gaussian Mixture Model for clustering, what-if simulations, and the use of interactive data visualization tools. These techniques were used first to understand the main fuel efficiency drivers, then to understand the drivers of safety, and finally to understand the trade-offs between fuel efficiency and safety with respect to different driving styles. Our results show that significant gains can be achieved in terms of fuel efficiency by changing driving behaviors. Results from the regression and simulator show that average speed, acceleration events and maximum RPM are the 3 most important variables for fuel efficiency. With small changes like increasing speed by 1km/h, reduce acceleration events in 5% and reduces maximum RPM by 5% fuel efficiency can be increased by 6%. We also demonstrate the main factors defining safety and their relative importance. Finally, we cluster driving styles and suggest good practices to replicate the best driving styles between different driving style clusters. Through a change management framework, we propose how some drivers could improve Coca-Cola FEMSA’s safety proxy by 34% without sacrificing fuel efficiency.