以地理加權邏輯斯迴歸探討國道嚴重事故因子

Using Geographically Weighted Logistic Regression to Investigate Factors Affecting Serious Crashes on National Highways

In recent years, government has been devoted to analyzing the frequency and severity of accidents on national highways. However, the rate of serious accidents still increases. Therefore, the study focused on A1 and A2 accident types and developed logistic regression and geographically weighted logistic regression to investigate the relationship between factors and severe highway crashes based on the data provided by National Highway Police Bureau. The results showed that“traffic flows”and“truck flow ratio”were negatively correlated with the probability of serious accidents, while“speed limit”,“multi-lanes”,“self-collision”,“night without light”,“vehicle speed”, and“poor mental state”were positively correlated. Moreover, the study demonstrated that significant variables had spatial heterogeneity, which meant there were spatial variance between variables and severe crashes. Additionally, based on AIC and performance indicators, it was found that regression with local estimators had better performance and classification results than regression with global estimators.