抄録
The objective of this paper is developing an accident prediction model at fourlegged signalized intersections in Seoul City to control random and local characteristics of accident. The first step is to classify and analyze the factors of accidents, and construct raw accidents data as an ordinal category. This step is able to make the structure of accidents data to include random characteristic, and the next step is to make a prediction model using GLIM (Generalized Log-linear models) including the error system having the negativebinomial distribution. Then, EB method (The Empirical Bayesian method) using the cross and time series data of subject elements is supplemented to the basic model, in order to correct the global prediction results. In this paper, the total 145 intersections in 156 intersections in Seoul are used, 80 for calibration and 65 for validation, and 11 intersections are abandoned because of short on data.
