Multidimensional risk assessment for vehicle trajectories by using copulas

Abstract

Fight against road unsafety is a French government priority. The policy conducted since 2002 allowed to obtain undeniable success. Despite this improvement, road accidents have very serious consequences on human level for road users. The proposed methods to reduce these accidents involve independence between criteria of accident risk. The objective of this study is to estimate the multidimensional risk of failure trajectory. It consists to investigate a new method of risk assessment in order to better characterize the dependence structure between the vehicle criteria for safety acceptance. This requires the use of simulation techniques such as copulas methods. This function connects the joint probability distribution to the marginal distribution. Thus, it contains all information on the dependence structure of models. However, the difficulty of multidimensional risk is to choose the copula which capture the better dependence between criteria. To select an adequate copula must be based on a statistical test. The χ 2 test used in the framework of the adjustment of a parametric distribution to an empirical distribution is in this regard an interesting tool for the choice of copula. The experiments have shown the relevance and effectiveness of this method. The results will help to better assess the risk of failure trajectory for vehicles.

Clustering using kernel entropy principal component analysis and variable kernel estimator

“…The Iris benchmark is one of the machine-learning datasets; Fisher first used it in [30]. It consists of 150 measurements of three distinct types of the Iris plant (Iris setosa, Iris virginica and Iris versicolor) of the four variables: width and length for sepal and petal.…”

“…It consists of 150 measurements of three distinct types of the Iris plant (Iris setosa, Iris virginica and Iris versicolor) of the four variables: width and length for sepal and petal. It is worth mentioning that, one of the classes is linearly separable, whereas the two others are not [30]. Considering the combination of kernel data transformation (mapping) and the EPCA as a preprocessing for input data, the figure presents results obtained on Iris dataset, Figure 3(a) illustrates the reduced features in the high-dimensional space (150 features).…”