Information Technology Reference
In-Depth Information
Detection and Automatic Signalling (MIDAS) system, which displays real-time traf-
fic information, designed to set variable message signs and advisory speed limits with
little human intervention. The sensors have been installed on major motorways and
some 'A' roads in the UK by the Highways Agency to monitor traffic conditions and
minimise congestion. The sensors are placed every 500m along a road and measure in
intervals of 60 seconds the number of vehicles on each lane (also known as flow),
their average speed, average headway (distance between vehicles with respect to
time) and the occupancy or the amount of time the vehicles spent on the sensing loop.
This project considers parts of the M4, A470 and M48 in the South of Wales.
Intuitively, traffic flow variables such as the number of vehicles on the road, their
speed and the distance between vehicles influence the risk of an accident happening.
Garber and Ehrhart [8] have found a relationship between the accident rate on two-
lane highways and the speed, flow and lane width. Golob et al. [4] investigated the
relationship between traffic flow, weather and lighting conditions and the accident
rate on urban freeways in California and have shown that in congested traffic flow,
both the risk of accident and its cost are correlated to the traffic speed and density.
Research has been carried out to use the traffic flow conditions prior to an accident
to predict the probability or risk of an accident happening in real-time. Golob et al. [5]
have identified temporal flow and speed patterns that are indicative of conditions that
could result in an accident and can be used to predict accidents in real time. Cheol et
al. [9] used the mean and standard deviation of flow attribute data over 5 minutes
before an accident to predict in real-time when current traffic conditions become ha-
zardous using a Bayesian modelling approach. A study by Lee and Abdel-Aty [7] also
used 5 min of traffic flow data before an accident to identify the traffic conditions
contributing to crashes on freeway ramps in Florida. However, most work done so far
focussed on temporal data rather than spatio-temporal data, which represents an added
level of complexity. Spatio-temporal data, in which data vary in both space and time,
from sensors have been used in a variety of applications including scientific domains,
including weather prediction and climate modeling, earth systems science, biomedi-
cine, and materials science [10, 11]. According to Treiber et al. [12] highway traffic,
cannot be adequately understood by time-series data alone but should be described
instead by considering data both with respect to time and space. In our work, we use
both temporal and spatial flow data to predict accidents based on historical spatio-
temporal flow data using case-based reasoning.
2.2
Case-Based Reasoning
In case-based reasoning (CBR), problems are solved based on the solutions of similar
past problems [3]. The case base consists of past cases, which are stored along with
their solution. Given a new case or target case, the CBR system calculates the similar-
ity between the new case and each case in the case base and then retrieves the most
similar case. The similarity measure, which calculates the similarity between the tar-
get case and the archived cases, is fundamental to the working of a CBR system. The
popular
K
-nearest neighbour method (KNN) [13] matches each attribute in the target
case to its corresponding attribute in the archive case. Attributes relevant to the
