Page 36 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

Spatio-Temporal Data Mining Analysis to Extract Association Rules Among Traffic
Accidents and Environmental Reports

1
1
1*
Zahra Irandegani , Mohammad Taleai , Reza Mohammadi

1 Faculty of Geodesy and Geomatics Engineering, K.N.Toosi University of Technology
Tehran, Iran
* taleai@kntu.ac.ir

Abstract: The 311 non-emergency service (also known as environmental reports) which is developed based on the
public participatory geospatial information system (PPGIS) has considered as a big data source including a huge number
of reports with a large variance in the form of location, time and types of reports. This study aims to explore the spatial
and temporal relationships between traffic accident reports, land use diversity and 311 service reports. For this purpose,
the Apriori algorithm is a data mining method utilized to extract association rules from accident data, land uses and
environmental reports. The dataset includes crash reports, the 311 reports, land-use parcels and road network for Boston,
gathered during three years 2015 to 2018. The results prove that there is a highly significant correlation between traffic
accidents and some citizen’s reports such as street cleaning, highway maintenance, street lights, sign and signal, sanitation
and enforcement & abandoned Vehicles. Moreover, a significantly strong relation between crash characteristics and land
use types could be demonstrated. For instance, Motor Vehicle accidents generally happened in the vicinity of residential
and commercial areas, and road intersections. Furthermore, most pedestrian crashes often happened in the street, during
the weekday, and in the proximity of the residential areas. This kind of results could be used to predict the locations that
are prone to crash events after receiving some related environmental reports. The potential of the presented association
rule mining results could be used in the decision support systems for traffic safety management.
Keywords— Big Data, Spatio-Temporal Analysis, Data Mining, 311 Non-Emergency Service, PPGIS, Apriori

1. Introduction
Public Participation Geographic Information System (PPGIS) provides some tools for citizens to take part in the decision-
making process [12] in local development projects. Recently, “311 non-emergency service” has published based on the concepts
of volunteered geographic information (VGI) data that enables citizens to report their local non-emergency problems ([14], [20]).
The spatial and temporal analysis could be used to extract useful information about neighborhoods from collected reports.
Machine learning algorithms (such as association rules) has emerged as a powerful gadget to release the latent pattern embedded
in these environmental reports.
Recently, some studies employed association rule mining to predict the severity of traffic accidents or the crash-pedestrian
injury. Geurts et al. (2003) used the association rule mining algorithm to identify accidents’ hot spots [8]. The results showed
that human and behavioral aspects are important in the frequency of accident patterns. Mutter et al. (2004) explored the use of
classification performance as a metric to evaluate the results of confidence-based association rule miners [18]. Mennis and Liu
(2005) applied multiple-level association rule mining to determine the spatial and temporal relationships based on a set of
variables that characterize socio-economic and land cover change in Denver, Colorado, U.S.A. during 1970 – 1990. Pande and
Abdel-Aty (2009) used association rule mining to investigate the relationship between different types of crashes including road
type, pedestrian age, lighting conditions, vehicle type [19]. Montella et al. (2011) and Montella et al. (2012) used classification
trees and association rules to detect interdependence and dissimilarity crash characteristics ([15],[16]). Das and Sun (2014)
applied the association rule mining technique to discover hidden patterns in rainy weather crash data of Louisiana [4]. The results
prove that in rainy weather, car accidents are influenced by some factors such as street lights at night, roadways alignment, day
of the week, and driver gender and age. Moradkhani et al. (2014) used the association rule mining technique to determine the
effective factors (such as light conditions, weather conditions, road surface condition, vehicle type, age of the driver, road type)
on road traffic accidents [17]. Tayeb et al. (2015) used the Apriori algorithm and explored the correlation between recorded
accidents’ factors and accident severity in Dubai [7]. Kumar and Toshniwal (2015) used a clustering algorithm followed by a
rule mining technique to predict locations that are prone to car accidents [11]. Das et al. (2018) used association rule mining to

1