Automotive ECU Data-Based Driver’s Propensity Learning Using Evolutionary Random Forest

Abstract

Driving assistance systems in the automotive industry are constantly evolving and are already commercialized in various areas to provide consumers with safety and convenience. The recognition of driver's propensity is a key factor that can greatly affect the performance of such a driving assist system, but it still has numbers of technical limitations. This paper presents an evolutionary machine learning algorithm for recognizing driver's propensity by effectively learning a vast amount of ECU sensor data in the vehicle, and its performance is verified through system construction, data collection, analysis, and comparison test. The experiments showed that the proposed algorithm achieves a classification accuracy of 92.48% in a large amount of ECU data and reaches 7.03% higher accuracy than the average classification accuracy of existing classifiers. In addition, a scenario for a new safe driving assistance system is presented. The system can recognize the driver's propensity in real time using only the ECU information without attaching additional sensors, such as cameras and biometric information. It is expected that this system will help to recognize the driver's tendency shift, thereby inducing safe driving.

Keywords

• Big data learning
• evolutionary computation
• machine learning
• evolutionary random forest
• driving propensity recognition
• vehicle safety assistant systems