Comparison of different hyperparameter optimization methods on driving behavior recognition

The prediction and recognition models of driving behaviors are often based on ma- chine learning approaches. These models are required for the growth of advanced driving assistance systems. The performance of the model depends on the optimal parameters, hy- perparameters, and model structure. In the present study, hyperparameters of a previously developed model (neural network-based state machine model) are optimized for the lane changing recognition. Two methods are considered for the hyperparameter optimization: Bayesian optimization and Genetic algorithm (GA). Three lane changing behaviors are estimated. Real human driving data generated using a driving simulator are used for the parameterization. The aim is to compare the model’s recognition performance based on the two methods. Furthermore, comparisons between the models with optimized hyper- parameters and the original model (without hyperparameter optimization) are performed. The results show that the performance based on the Bayesian optimization is better than GA, while the original model still outperforms others.