The Employment of Artificial Neural Network in Predicting Malaysian Road Traffic Deaths Towards 2030

Developing a reasonable prediction of the number of fatalities from road accidents could assist stakeholders in creating effective policies to reduce these deaths. An exponential model using historical Malaysian road accident data from 1972 to 1994 was previously developed to predict fatalities. However, this model failed to reasonably predict fatalities after 1998, likely due to the nonlinearity of the data. The present study proposed an artificial neural network (ANN) model to predict Malaysian road deaths using the same parameters as the exponential model. A single-hidden-layer ANN with a 4-10-1 topology was developed, using hyperbolic tangent sigmoid activation and the Levenberg-Marquardt learning algorithm. The ANN model provided better prediction than the exponential model, with a mean absolute percentage error of 19%. The model forecasted 8,610 deaths in 2030. It is evident from the study that non-conventional statistical techniques like ANN can provide reasonable predictions of Malaysian road deaths whilst adding relevant features could further improve the predictability of the model.