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.

Keywords: MDLSTM; interpretation; spatial; traffic accident.

Figure 1

(a) The spatial distribution of the “traffic accident counts” indicator; (b) The spatial distribution of real traffic accidents.

Figure 2

(a) The distribution of accident date. (b) The distribution of accident time. AM and PM in (b) are “ante meridiem” and “post meridiem” which means time period before noon (0:00–12:00) and after noon (12:00–24:00).

Figure 3

(a) The spatial distribution of the isolation form. (b) The spatial distribution of the cross-sectional location.

Figure 4

(a) Schematic of the urban area in the City of Shenyang with roads. (b) The plot ratio data after rasterization.

Figure 5

The window sampling and batch splitting.

Figure 6

Application of recurrent neural network in natural language processing. “t” means the word step while “t − 1” means the previous step of “t”. The “h(t)” means the output of step “t” and the “x(t)” means the input of step “t”. The “?” means the word that needs to be predicted corresponding to the predicting result word “Lied”.

Figure 7

(a) The model structure of MDLSTM. (b) The application in the accident analysis. ”t” and “s” shows the coordinate location of the grid cell. For example, the (t − 1, s) is the left grid cell of (t, s). ”A” in this figure shows the MDLSTM cell in a special location, for example the “A” in the tth column and sth row corresponding to the MDLSTM cell in the same location. The blue arrow means the information transfers from the neighboring grid cell.

Figure 8

The structure of MDLSTM cell. Arrows in the figure shows the linear transformation from the variable behind the arrow to the variable in front of the arrow.

Figure 9

Performance of MDLSTM, LSTM, RNN and BPNN on the training dataset. MDLSTM means the “multi-dimensional long-short term memory neural network”. LSTM means the “long-short term memory neural network”. RNN means the “recurrent neural network”. BPNN means the “back-propagate neural network”.

Figure 10

The spatial distribution of state value $C_{t,s}$.

Figure 11

(a) The spatial distribution of potential of accident count. (b) The spatial distribution of date away from winter.

Figure 12

(a) The spatial distribution of the potential of accident time. (b) The spatial distribution of the isolation form.

Figure 13

The values of intermediate variables coming from 30 continuous windows in a batch behind the grid cell (50, 40). Each color of the figure shows the corresponding value of the type of intermediate variable. For example, the “i − 1” variable shows the first element of the i_t,s.