Predicting Road Accidents Based on Current and Historical Spatio-temporal Traffic Flow Data - Computational Logistics - page 88

t m and d n represent the sensor indices just prior to the point at which the accident risk
is to be assessed, whereas t 1 and d 1 represent the first sensor from which onward sen-
sor measurements are considered with respect to time and distance, respectively. The
case base consists of historical accident cases and non-accident cases along with their
corresponding flow data. The flow data in the target case is compared to the flow data
corresponding to each accident case and their similarity is calculated. The most simi-
lar case is retrieved. If the outcome of the retrieved case, Outcome (C C ), i.e. either
accident or non-accident case, is the same as the outcome of the target case, Outcome
(C T ), the retrieval is deemed as successful. This procedure is repeated for all accident
and non-accident cases as target cases. The average success rate gives an indication of
the capability of the system to differentiate between accident and non-accident cases.

3.3 Case Attributes

The knowledge contained in a CBR system is represented in the form of cases. The
case attributes describe a case using information that is relevant to the solution of the
CBR system. In the CBR system under development the cases comprise the traffic
flow data over a certain period of time and a certain distance.
The target case contains the current traffic flow data or the flow data for which we
want to determine the accident risk. The cases in the case base represent historical
accidents and the traffic flow conditions in which they occurred or non-accident cas-
es. The solution of a case is the classification, i.e., has an accident occurred or not.
The case attributes are stored in the case base in the form of key value pairs. The
attributes can be divided into four groups:

1. Nominal attributes

These attributes take categorical values. They are determined for the entire spatio-
temporal period of flow in consideration. They include:

• Time day category: The time of day attempts to capture the traffic state and there-

fore congestion that can be expected. This attribute can take six nominal values:

─ '1' - represents the morning rush hour on a weekday and has been empirically set
between 6:00h and 10:00h from Monday to Friday
─ '2' - represents normal traffic flow between 10:00h and 16:00h on weekdays from
Monday to Friday.
─ '3' - represents the evening rush hour traffic state between 16:00h and 19:00h
from Monday to Friday
─ '4' - represents night time traffic between 19:00h and 6:00h.
─ '5' - represents weekend (Saturday/Sunday) day time traffic between 6:00h and
19:00h.
─ '6' - represents weekend night time traffic between 19:00h and 6:00h.

• Time-interval/Distance, T/L : refers to the period of time T or distance L over

which flow data is considered. It is currently an empirically set design choice.