Республиканская научно -техническая конференция транспорт-логистика: проблемы и

 
 
 

---

46 
PREDICTING TRAFFIC ACCIDENT HOTSPOTS USING LOGISTIC 
REGRESSION 
 
DSc. Sadikov J.I., Associate Professor, Imamaliev D.M. Doctoral student,
Jalolov S.H Master's student (TSTU) 
 
 Introduction. Presently, more than 1.35 million people are killed every year 
worldwide as a result of road accidents caused by cars; meanwhile, road accidents have 
become the main cause of death and pose a threat to people's lives [1]. The occurrence 
of road accidents can be explained by several factors but not limited to, such as drivers’ 
behavior, vehicles and roads condition, the environment, etc. Therefore, if it is possible 
to find out the relationship between road accidents and various factors, the probability 
of accidents can be effectively reduced by early warning before an accident, and this 
will be very useful for road safety. The paper will examine data on a number of road 
accidents on the roads. Proposed method uses 6 independent variables which 
characterize people’s age? vehicles type? and the environment? Using logistic 
regression, the relationship between an accident and several factors should be analyzed, 
and then a model for predicting traffic accident hotspots will be created, which would 
provide a theoretical basis for traffic accident prevention and active vehicle safety 
technology. 
Analysis.Data on road accidents should include detailed statistical information 
about road accidents, such as: the place of the accident, the slope of the road, the time, 
the cause of the accident, the state of lighting, road condition, the number of fatalities, 
etc. As a rule, a driver is considered as a subjective factor, while the road, vehicle, and 
environment are objective factors. Therefore, when choosing risk factors for a road 
accident, it is necessary to take into account not only subjective factors, but also 
objective ones. In this study, 6 independent variables with different aspects were 
selected, as shown in Table 1 [2]. 
Suppose that there are n factors that affect the occurrence of accidents, denoted 
by: x
1
, x
2
, ..., x
n
. The logistics model is as follows [3]: 
log 𝑖𝑡 (𝑦) = ln (
𝑝
1−𝑝
) = 𝑎
0
+ 𝑎
1
𝑥
1
+ 𝑎
2
𝑥
2
+. . . +𝑎
𝑛
𝑥
𝑛
(1) 
Where y = (0, 1) is an estimate of the probability of a road accident; p is the 
probability of a road accident; x
i
(i = 1, 2,..., n) is the factor associated with the road 
accident; a
0
is the constant, a
i
(i=1, 2,..., n) are the regression coefficients. 
The model can be equivalently expressed as [4]: 
𝑝 =
𝑒
(𝑎0+𝑎1𝑥1+𝑎2𝑥2+⋯+𝑎𝑛𝑥𝑛)
1+𝑒
(𝑎0+𝑎1𝑥1+𝑎2𝑥2+⋯+𝑎𝑛𝑥𝑛)
(2)