This article aims to address the challenge of recognizing driving styles, a task
that has become increasingly complex due to the high dimensionality of driving
data. To tackle this problem, a novel method for driver style clustering, which
leverages the principal component analysis (PCA) for dimensionality reduction
and an improved GA-K-means algorithm for clustering, is proposed. In order to
distill low-dimensional features from the original dataset, PCA algorithm is
employed for feature extraction and dimensionality reduction. Subsequently, an
enhanced GA-K-means algorithm is utilized to cluster the extracted driving
features. The incorporation of the genetic algorithm circumvents the issue of
the model falling into local optima, thereby facilitating effective driver style
recognition. The clustering results are evaluated using the silhouette
coefficient, Calinski–Harabasz (CH) index, and GAP value, demonstrating that
this method yields more stable classification results compared to traditional
clustering methods. In the final stage, a particle swarm optimization-SVM
(PSO-SVM) algorithm is applied to classify the clustering results, which are
then compared with results from other machine learning algorithms such as
decision tree, naive Bayes network, and K-nearest-neighbor (KNN). This
comprehensive approach to driver style recognition holds promise for enhancing
traffic safety and efficiency. The accurate recognition of driving style can lay
the foundation for further optimization of advanced driver assistance systems
(ADAS).
