Clustering and prediction of the driving style in the cut-in process

• Main Author: Xiao, Hongzhao
• Other Authors: Su Rong
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling
• Online Access: https://hdl.handle.net/10356/173231
• Institution: Nanyang Technological University

For a long period, autonomous vehicles (AVs) and human-driven vehicles (HDVs) need to share roads in mixed traffic flow. The cut-in behavior of HDVs represents a common and hazardous traffic action faced by AVs, and this behavior may occur frequently. To better understand and address the cut-in behavior, understanding and analyzing the driving style associated with this behavior are crucial. This dissertation investigates how to classify, analyze, and predict the driving style of the cut-in process. The features of the driver behavior are selected from the speed-change and lane-change phases of the cut-in process. The principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) methods are employed to reduce the dimensionality of the features. The k-means++ algorithm is applied to cluster the driving style of the cut-ins. Driver-in-the-loop experiments were conducted to acquire cut-in data with different driving styles. The clustering results indicate that optimal clustering performance is achieved when employing the t-SNE method with a cluster number set to three. Based on the clustering results, a statistical analysis was conducted to illustrate the characteristics of three different cut-in driving styles. Subsequently, deep learning methods, along with other machine learning approaches, were applied to predict the cut-in driving styles of HDVs. Experimental results suggest that prediction models using a time window approach exhibit a certain level of prediction accuracy.