Determination of defect rates of traffic accident involvements by using accident reconstruction tools

The aim of this study is to introduce a scientific and systematic approach for determination of fault rates in most frequent traffic accidents in Turkey. Data (police reports, skid marks, deformation situation of involvements, crush depth, etc.) collected from the most-frequent and controversial accident types (four sample vehicle-vehicle scenarios), were inserted into a reconstruction software called "vCrash". Sample real world scenarios were simulated on the software so as to generate different deformations on vehicles which also correspond to energy equivalent speed (EES) data just before the crash. These values were used to train Multi-Layer Feed Forward Artificial Neural Network (MFFNN or MFANN), Function Fitting Neural Network (FITNET) (i.e. a specialized version of MFFNN), Generalized Regression Neural Network (GRNN) by using 10-fold and 5-fold cross-validations. Two approaches within ANN were used to predict fault rates as accurate as possible without necessity of softwares. The performance of Artificial Neural Network (ANN) prediction models was evaluated using Mean Square Error (MSE) and multiple correlation coefficient (R). MFFNN and FITNET perform better results (i.e., lower MSE and higher R) than GRNN models for predicting the fault rates.