Examining the influence of rural land uses and accessibility-related factors to estimate pedestrian safety: The use of GIS and machine learning techniques

Land uses are of paramount and practical importance when it comes to populated rural areas. In spite of this fact, not much research has been conducted to discover the interaction between roadside land uses, and pedestrian crashes in rural areas. As a result, the present study tries to explore unsafe segments of populated rural areas for pedestrians with emphasis on roadside land uses variables and access road locations to address this gap. To achieve the objective of this research, logistic regression as a parametric and classification and regression trees (CARTs) as a nonparametric method were used for estimating unsafe segments of the studied road based on different variables combinations, regarding their correlation. Separate models were developed for each combination, and the selection between the models was conducted by goodness-of-fit measures and the model's accuracy and comprehensiveness. The result of testing the proposed approach on historical crash data of several segments of a rural multilane road demonstrates that the logistic regression method provides slightly superior results compared to the CART in terms of unsafe segment prediction for pedestrians. In line with expectation, residential land uses are found to have the worst effect on the safety situation of study road segments. Besides, the results show that commercial and retail, governmental, and institutional land uses, together with the number of access roads, also adversely affect pedestrian safety on rural multilane roads. After discovering the effective land use-related variables, using geospatial analysis, the predicted safe and unsafe segments are compared with the real safety condition of segments based on historical crash data.