Predicting and Interpreting Spatial Accidents through MDLSTM

Abstract

Predicting and interpreting the spatial location and causes of traffic accidents is one of the current hot topics in traffic safety. This research purposed a multi-dimensional long-short term memory neural network model (MDLSTM) to fit the non-linear relationships between traffic accident characteristics and land use properties, which are further interpreted to form local and general rules. More variables are taken into account as the input land use properties and the output traffic accident characteristics. Five types of traffic accident characteristics are simultaneously predicted with higher accuracy, and three levels of interpretation, including the hidden factor-traffic potential, the potential-determine factors, which varies between grid cells, and the general rules across the whole study area are analyzed. Based on the model, some interesting insights were revealed including the division line in the potential traffic accidents in Shenyang (China). It is also purposed that the relationship between land use and accidents differ from previous researches in the neighboring and regional aspects. Neighboring grids have strong spatial connections so that the relationship of accidents in a continuous area is relatively similar. In a larger region, the spatial location is found to have a great influence on the traffic accident and has a strong directionality.

Identification and Prioritization Techniques Applied to Urban Accident-Prone Locations: A Scoping Review

“…Fourteen grouping methods were found among the 42 studies evaluated (Table 3), highlighting the use of section split (SS) (Al-Omari et al, 2020;Greibe, 2003;Guerrero-Barbosa and Santiago-Palacio, 2016;Halim et al, 2018;Hayidso et al, 2019;Nguyen et al, 2016;Qu et al, 2019;Saha et al, 2020;Sugiyanto et al, 2017;Yan et al, 2019), kernel density estimation (KDE) (Hegyi et al, 2017;Le et al, 2020aLe et al, , 2020bPleerux, 2020;Shafabakhsh et al, 2017;Soltani and Askari, 2014;Xie and Yan, 2013), network kernel density estimation (NKDE) (Al-Aamri et al, 2021;Fan et al, 2018;Nie et al, 2015;Xie and Yan, 2008), spatial autocorrelation (SA) (Chance Scott et al, 2016;Steenberghen et al, 2011;Ulak et al, 2019), community census (CC) (Dong et al, 2016;Dumbaugh et al, 2010;Vaz et al, 2017), and cells (Cui and Xie, 2021;Debrabant et al, 2018;Geurts et al, 2005;Xiao et al, 2021). Less frequently used were beta-binomial screening (Park and Sahaji, 2013b), city limits , DBSCAN (Szénási and Jankó, 2017), the Firefly algorithm (Yuan et al, 2020), Gaussian mixture models (GMMs) (Mansourkhaki et al, 2017), nearest neighbor (NN) (Rahman et al, 2020)…”

“…Four studies used census community limits to define spatial blocks (Dong et al, 2016;Dumbaugh et al, 2010;Vaz et al, 2017). Four other studies used spatial cells defined from raster cell geometries or hexagons of equal dimensions, forming equidimensional neighborhood spatial units (Cui and Xie, 2021;Debrabant et al, 2018;Geurts et al, 2005;Xiao et al, 2021).…”

“…Variation in the spatial coverage of the UAs was also identified, with applications at the macro level Dong et al, 2016;Xiao et al, 2021), the meso level (Dumbaugh et al, 2010;Vaz et al, 2017), and the others at the micro level.…”

“…This strategy enabled the evaluation of spatiotemporal correlations using after-before analyses (Cui and Xie, 2021;Geurts et al, 2005). The exact cell cutout could be used in the intersection of cell areas with the occurrence of RTAs and associated risk factors, allowing the estimation of potential accidents as a function of factors contributing to neighborhood characteristics (Debrabant et al, 2018;Xiao et al, 2021).…”

“…In the SA method, the behavior of APL candidate sites is described in terms of neighborhood relationships, where correlations between measures of local characteristics and physical events enable understanding of the ways that contiguity relationships influence local phenomena. Such approaches aim to understand shared effects determined by factors such as the nearby environment , spatial density (Hegyi et al, 2017;Soltani and Askari, 2014;Steenberghen et al, 2011), temporal variation (Chance Scott et al, 2016Fan et al, 2018;Le et al, 2020a), expected frequencies (Cui and Xie, 2021;Dong et al, 2016;Xiao et al, 2021), contributing frequencies (Geurts et al, 2005;Vaz et al, 2017), and frequencies above the upper control limit (Pleerux, 2020). SA can also be used to estimate effects resulting from mitigation measures, where the frequencies suppressed by mitigation measures are discounted from the expected values (Le et al, 2020b).…”