Application of the Neural Network and Structural Time Series Models 2002-01-2007
For many years, time series analysis methods have been applied to develop models linking serious road crashes with the effects of socio-economic and other factors. Classical time series models, such as log-linear regression or ARIMA, are useful in describing observed road trauma but are typically poor when used for forecasting road safety trends. In recent years, a new method of forecasting has been developed called an artificial neural network (ANN). The main focus of this research was on examining the possibility of using ANNs for prediction and to find complex relationships ‘hidden’ in time series models that classical modelling techniques often fail to reveal. Application of structural time series models for this purpose was also investigated.