Forecasting the Trend, Seasonal and Random Components of the Average Accident Rate on Public Roads Using Recurrent Neural Networks

The paper deals with investigating the ability to make a forecast of the average level accident rate on the public roads by means of recurrent neural networks. For increasing the forecast accuracy we use time-series decomposition (corresponding to the average level accident rate) into trend, seasonal and random components. The paper describes the architecture of a cascading neural network making a forecast based on the additive time-series decomposition. The accuracy of the obtained forecast is assessed.