Abstract use machine learning Artificial Neural Network and

Abstract

In today world many researches and engineers are
working on to make more protected and safer automobiles, but still the traffic
accidents are unavoidable. There is pattern of dangerous crashes that could be
classify and can be helpful in designing and developing accurate
automobiles.  These patterns are very
useful for road safety and for traffic police departments. In zone-level crash
estimation, accounting for latitudinal dependence has become an extensively
studied Topic. A transportation network is a collection of
road-traffic-environment modules and features multi categories of
interdependent factors. This mix makes the organization of safety in traffic
study zones explicitly challenging. This study suggests Support Vector Machine
(SVM) model to address complex, large and multidimensional spatial data in
crash prediction. In this research I am going to use machine learning
Artificial Neural Network and support vector Machine learning techniques to
predict next occurrence of accident.

 

Keywords

Automobiles crashes, Artificial Neural Networks (ANN),
road traffic accident, machine learning, support vector machine, Traffic
accident data mining, road accident severity prediction, road accident sensitivity
analysis, and performance comparison.

Introduction

The expenses of fatalities and wounds because of
traffic accident greatly affect the general public. In later a long time,
specialists have given careful consideration to deciding elements that
fundamentally influence seriousness of driver wounds caused by auto collisions.
There are a few methodologies that analysts have utilized to consider this
issue. These incorporate ANN, SVM, and Bayesian.

Many applications have been developed to assess safety
level of several types of road entities and to examine effect of safety
countermeasures. Crash prediction model (CPM) is an important tool in traffic
safety analysis. Freshly, traffic crashes are collected by a certain spatial
scale and researchers usually seek to relate safety to zone-level factors. The
main objectives of macro-level crash prediction analysis are to explain
observed cross-sectional variations in safety using zone-level covariates at
different spatial scales. These macro-level CPMs may aid in more effectively
joining safety consideration into transportation planning and management.

The Correlation-based Feature Selector (CFS) was
applied to evaluate candidate factors possibly related to zonal crash frequency
in handling high-dimension spatial data. To determine the future approaches and
to compare them with the Bayesian spatial model with provisional autoregressive
prior. A dataset in Hillsborough County of Australia was employed. The results
showed that SVM models accounting for spatial closeness outclass the
non-spatial model in terms of model suitable and predictive performance. The
best model predictive capability, while further exhibits SVM models’ capacity
for addressing moderately complex spatial data in local crash prediction
showing. SVM models are better than CAR models. A sensitivity analysis of the
centroid-distance-based spatial SVM models could capture the impacts of
explanatory variables on the mean predicted chances for crash occurrence. While
CAR models, which supports the employment of the SVM model as another in local
safety modeling.

 

Related Work

Here has been some work in the field of road anomaly
detection.

Applying data mining systems to show auto collision
information records can comprehend the qualities of drivers’ conduct, roadway
condition furthermore, climate condition that were causally associated with
distinctive damage seriousness. This can help leaders to plan better movement
wellbeing control strategies. A researcher has worked how factual strategies in
view of coordinated charts, developed over information for the current period,
might be helpful in displaying movement fatalities by contrasting models
determined utilizing coordinated diagrams with a show, in view of out-of-test
estimates, initially created by Peltzman.

Another researcher Ossenbruggen et al utilized a
strategic relapse model to distinguish measurably critical factors that foresee
the probabilities of accidents and damage crashes going for utilizing these
models to play out a hazard evaluation of a given district. These models were
elements of elements that portray a site by its property utilize action,
roadside configuration, utilization of movement control gadgets also, activity
presentation. Their examination outlined that town destinations are less unsafe
than private and shopping destinations. Abdalla et al. examined the connection
between loss frequencies and the separation of the mischances from the zones of
home. As might have been expected, the setback frequencies were higher closer
to the zones of home, perhaps because of higher presentation. The investigation
uncovered that the loss rates among occupants from territories delegated
generally denied were altogether higher than those from moderately prosperous
zones.

Another researcher Miaou et al examined the measurable
properties of four relapse models: two regular direct relapse models and two
Poisson relapse models as far as their capacity to show vehicle mischances and
parkway geometric plan connections. Roadway and truck mischance information
from the Highway Security Information System (HSIS) have been utilized to
represent the utilization and the impediments of these models. It was exhibited
that the traditional straight relapse models do not have the distributional
property to depict enough arbitrary, discrete, nonnegative, and commonly
sporadic vehicle mischance occasions out and about. The Poisson relapse models,
then again, have the vast majority of the attractive factual properties in
building up the connections.

The paper 1 the system and the procedure knowing the
conditions of the roads surface. They used GPS and external accelerometer, the
sensor prepared vehicles. After collecting the data by GPS sensor and
vibration. After that, that data help to know the condition of road surface.
They have arranged the system on 7 taxis running in the Boston area.

2 This Research developed by Microsoft. It uses
multiple external sensors such as a microphone, GPS, accelerometer for
detection of ruts, and Global System for the location of potholes. It helps us
to detect the traffic honking, bumps etc. by using different sensor.

In paper 3 If any accident or trouble happens at a
certain place. That situation immediately inform to the relative. For that
purpose, they have used a wireless short message service. This device consists
of the collision sensors, micro controller unit, GPS and GSM. That data
obtainable can educate the driver about the road conditions. That help in
travelling.

In other research 4 For knowing the irregular
condition, they have used a android os based smart phones. The evaluation
presented with true positive rate is 90 % using real world data.

The author of paper 5 to measure a vehicles
conditions, some of them are gear shifts likes and overall road conditions
which consist of bumps, rough road, uneven road, and smooth road. For that Safe
Driving Purpose Using Mobile Phone. In this they mentioned some advanced
applications which integrated inside an automobile to assess a vehicles
conditions.

In paper 8 for prediction of bumps on the roads.
They have used accelerometer and the GPS of smart phone feature.

In paper 6 describes a pothole detection system. The
neural network technique is used for justifying the threshold values. This
approves an accuracy of 90 %-95%. The paper 9 to find the t the types of
roads and extracts effective features from different vehicles. The machine
learning procedure. a training data set is used to find the roadway types.
Machine learning algorithm is applied for this purpose.

Road Accident death Dataset

In this research paper an official database of Australian
Road Deaths Database (ARDD) is used. The Australian Road Deaths Database gives
essential subtle elements of street transport crash fatalities in Australia as
detailed by the police every month to the State and Territory Street wellbeing
specialists.

This dataset contain 48366 records from 1989 till 2017.
Each record contain the following detail of road accident.

·        
Crash ID

·        
State

·        
Date

·        
Month

·        
Year

·        
Time

·        
Crash type (Multiple, Pedestrian, Single)

·        
Bus Involvement (Yes, NO)

·        
Heavy Rigid Truck Involvement (Yes, NO)

·        
Articulated Truck Involvement (Yes, No)

·        
Speed Limit (from 0 to 900)

·        
Road Users (Bicyclist, Driver, Motorcycle
pillion passenger, Motorcycle rider, Other/unknown,
Passenger,
Pedestrian)

·        
Gender

·        
Age

Dataset source is

https://data.gov.au/dataset/5b530fb8-526e-4fbf-b0f6-aa24e84e4277/resource/0e3771a4-4783-4a12-89ac-b48892bb3ba0/download/fatalcrashesoctober2017whv.csv

 

 

Methodology

Support vector machine:

Support vector machine (SVM) has very important role
in classification of machine learning classification algorithm. This algorithm
is mostly used to classify different classification problems. This algorithm is
basically designed based on statically theory and risk reducing problems.

This algorithm basically draw or find hyper plan
between classifications to final target boundaries. For this study I have classification
of Crash Type. That is Multiple (represent with 1) , Pedestrian ( represent
with 2) and Single ( represent with 3).

(x1, y1,z1),. . .,(xi, yi,zi), yi ? {1, 2,3}  

Written by