Driving Style Prediction Using Clustering Algorithms

Behavior prediction of surrounding vehicles is a critical task. The main goal of this work is the implementation of Gaussian Mixture Model (GMM) and K-means to predict behavior of other vehicles moving on the road and theoretical analysis of their performance. All the vehicles are clustered in three different clusters (aggressive, moderate and conservative) depicting their driving styles. In order to achieve this goal, we implemented GMM and K-means with the help of keras in Python 3.6. The models are tested with NGSIM (Next Generation Simulation) data on the US-101 and I-80 dataset. The results of both the algorithms are visualized using scatter plots. Statistical properties of driving styles are derived using the statistical properties of records belonging to that cluster. The performance of both the algorithms is compared.