Driving Behavior Improvement and Driver Recognition Based on Real-Time Driving Information

Based on real-time driving information recording data, we raised a regression model for predicting gasoline consumption rate from speed, acceleration and heading degree information. We also developed a grouping method to sort high-frequency fragmented driving information to sets of trips, and then implemented supervised learning to illustrate the relationship between average speed and gasoline usage. The general MPG-MPH relationship gives useful suggestions of good driving speed. The ranking of different vehicles’ MPG from real-time driving information are more close to practical data, compared with the data given by automakers, which is a useful criteria for new vehicle consumers. We also implemented both supervised learning (Näıve Bayes and Support Vector Machine) and unsuperivised learning (k-means clustering) to try to classify potential multi-driver vehicles. Though the ultimate goal is still not achievable, we can still classify driving behaviors to 2 patterns with different speeds, which is a hint for distinguishing different driving conditions, and driver’s driving preference, which could be a good foundation for further work.