accidents, thus promoting road safety. Considering the problems of time and money, it is not practical to

improve road safety in all the places where road traffic accidents occur. Therefore, the process of identifying

accident-prone locations, known as black spots, is a cost-effective and efficient way to analyze the causes of

road accidents and reduce them. Identifying black spots is an effective strategy to reduce accidents. The core

methods that may be used in the process of identifying the black spots of a road network are the sorting,

grouping, and accident prediction methods. However, in practice, it is easy to overlook certain factors that

significantly contribute to defining and characterizing a spot on the road network as black. Therefore,

suggestions to carry out projects required to reduce security risks shall not be based on the above methods.

Machine learning algorithms that in recent years have been widely used in the field of predicting a road traffic

accident cover these weaknesses. They can effectively classify data sets and make a connection between factors

specific measures to determine the quality of a model, which resulted from the application of a logistic

regression and machine learning algorithms.

Key-Words: -Black spot, road safety, traffic accidents, binary logistic regression, data set, machine learning

Received: August 28, 2022. Revised: April 19, 2023. Accepted: May 5, 2023. Published: May 29, 2023.

1 Introduction

In modern times, the improvement of human living

the list of the most important social problems and

form an ever-present threat to all road users. They

are complex phenomena that researchers must deal

with [1], given that they involve the simultaneous

interaction of a multitude of factors. The most

plays the most decisive and important role, having

achieved through the development of methodologies

that enhance the identification of road network

locations in which an increased concentration of

accidents is observed and the risk of collision is

characterized as high, [15]. These specific locations

are called "traffic accident black spots" or

"dangerous road spots".

long-term countermeasures are suggested for their

restoration. Identifying them is a cost-efficient and

the road network (such as wrong slope, lack of street

lighting, incomplete vertical and horizontal

markings), but also driving behaviors (such as

distraction, excessive speed) that demonstrably lead

to traffic accidents.

to:

a) the lack of data and specifically the limited

after an autopsy on the accident scene. Copies of the

DOTA are sent to the Hellenic Statistical Authority

data with the impossibility of assumptions were

removed, as well as the unclear data such as street

for classifying a spot of the road network as black,

the threshold used by ELSTAT was adopted as the

threshold for the above classification, which is

nothing more than at least two accidents. By using

the four classic methods of determining black spots

(Poisson random distribution method, Quality

control method, Accident frequency method,

Severity index method) we confirmed, at a 90%

significance level, the black spot characterization, as

defined by the ELSTAT (two accidents at the same

with a predefined range of values, to reflect, as

appropriate, the various ELSTAT data through

where the input variable z represents the action of a

set of independent variables, and f(z) determines the

where β0 is the constant term (slope) of the

regression line, and βi are the regression

coefficients, expressing the magnitude of

contribution to the model of the corresponding

variable. A negative value of βi indicates that the

independent variable reduces the probability of an

event occurring, while a positive value of the

explanatory variable indicates an increase in the

4.2 Κ-nearest Neighbors

The most basic method of finding records that

belong to the same class based on distance is the K-

nearest neighbors’ algorithm. In this algorithm, we

assume that all instances correspond to spots in the

n-dimensional space Rn. The nearest neighbors of an

instance are defined by the Euclidean distance.

More specifically, let’s say that an instance x which

is described by the feature vector ((a1(x), (a2(x), …,

(an(x)) where ar (x) defines the value of the r-th

where V is the finite set {u1, u2…, us}, the algorithm

approximates a discrete-valued “target” f with the

estimate f

and k of the training set closest to x. The algorithm

can be divided into two parts:

(typically the Gini Index). This random sample of

features leads to the creation of multiple de-

Index is used in the construction of the forest) is

computed. This value is called the Gini Importance

of the feature. To perform feature selection, each

feature is ordered in descending order according to

the Gini Importance of each feature and the user

selects the top k features according to his/her

choice.

It is a neural network where the mapping between

inputs and output is non-linear. A Multilayer

Perceptron has input and output layers and one or

more hidden layers with many neurons stacked

together. While in the Perceptron the neuron must

have an activation function that imposes a threshold,

like ReLU or sigmoid, neurons in a Multilayer

Perceptron can use any arbitrary activation function.

feed-forward algorithms because inputs are

the data. This goes all the way through the hidden

layers to the output layer.

"Wet-Wet" and "Other" values, while there are

between 2 and 6. Thus, the distance between vectors

transformation can be used by the above algorithms

without questioning the correctness of their

application and the results. Figure 2 illustrates the

proposed method of this study in discrete steps.

machine learning classifiers.

generalized non-linear compression, [25]:

logistic regression was performed using the

Backward LR (likelihood ratio) method.

three metrics were computed for the evaluation of

models such as Accuracy, Recall, and F1 Score,

[26].

the fraction of total correct forecasted values

divided by the total number of test examples

(dataset):

Accuracy

FP: the number of values wrongly predicted as

FN: the number of values wrongly predicted as

positive values of the dataset were correctly

identified by the model:

Recall

is determined as:

Precision

precision and recall and calculates how much a

model perfectly classifies every value of the dataset

in the right category:

F1 Score

phenomenon called road traffic accidents requires

the collection of a large number of reliable data, as

lead to road traffic accidents and involve the

characteristics of the driver, the vehicle, and the

road environment. To solve the binary class

problem, the study ended up proposing a four-step

approach:

iii. extracting hidden space vectors, and

iv. training and using a classifier.

the individual factors that contribute to causing road

traffic accidents and the limited and at the same