GENERALIZED LINEAR AND GENERALIZED ADDITIVE MODELS IN STUDIES OF MOTORCYCLE ACCIDENT PREDICTION MODELS FOR THE NORTH-SOUTH ROAD CORRIDOR IN SURABAYA
Abstract
One of the advances in the development of a model of traffic accidents is indicated by the availability of generalized linear models (GLMs) and generalized additive models (GAMs) in the regression analysis. This paper will discuss the motorcycle accident prediction models using GLMs and GAMs on the north-south road corridor in Surabaya. The first part will discuss the model prediction of traffic accidents, as well as providing a brief review related to the use GLMs and GAMs in building models of accidents. Furthermore, application examples of GLMs and GAMs will be presented. To determine the effect of non-linear in each explanatory variable, smoothing the GAMs will be conducted in each variable gradually. Model diagnostic and intrepretations will be done in the final part. The results of the application of GLMs and GAMs indicates that the development of predictive models of motorcycle accidents with statistical methods can be used to diagnose problems in road safety. GAMs produces better models than the GLMs in which its condition without using the Poisson distribution, as shown in the difference in the value of the model parameter R-sq.(Adj), deviance explained, and the GCV score. By using the Poisson distribution with a log link-function, it appears that GLMs and GAMs produce the same model parameter values.