A Density-Based Clustering of Spatio-Temporal Data - New Contributions in Information Systems and Technologies - page 37

vehicles). The data generated by the moving objects are normally available in the
databases as sample points, in the form ( tid, x, y, t ), where tid is the object identity, x
and y are the spatial coordinates of the moving object at time t [3]. Handling a moving
object database is not a trivial task. It began decades ago but since its complexity was
relatively high, researchers first began looking into spatial data and temporal data
separately. Later on, study of spatio-temporal databases began and many models were
proposed [4].
Data mining of spatio-temporal data is the extraction of hidden predictive
information from large Moving Object Databases. It is a powerful technology with
great potential to help one focus on the most important information in their huge data
warehouses [5]. Clustering spatio-temporal data is one technique making it able to
analyze this data and discover similarity between trajectories of moving objects. It
groups the data according to their similarity into meaningful clusters. Data in each
cluster share common characteristics which could be defined many ways. For
example, objects in a cluster minimize the distance from the centroid of the group,
meaning they are close to each other, while at the same time maximize the distance to
objects in other clusters, making them dissimilar [6].
The main objective of this study is to provide a method to analyze spatio-temporal
data. This research area is believed to be very important because it could help make
future predictions about patterns of moving objects. This will therefore make decision
making easier or give possible answers to questions like, where will it end up and
when?
The rest of this paper is organized as follows. Section 2 presents background and
related work to the developed algorithm. Section 3 presents the developed algorithm
in this study. Section 4 represents the experimental evaluation and analysis of this
study. Section 5 discusses the experimental tests. Finally, section 6 concludes the
paper.

2 Methodology

TRACLUS is an algorithm that was suggested to cluster sub-trajectories instead of
clustering trajectories as a whole. Figure 1 represents the main sections of this
algorithm. TRACLUS consists of two phases: a partitioning phase followed by a
grouping phase [7].
The first phase is responsible for partitioning each trajectory into smaller segments
using minimum description length principle (MDL). The following phase is a phase
that groups the segments of all the trajectories into similar clusters using a modified
version of DBSCAN that works with lines segments instead of points. This modified
DBSCAN has two input parameters: Epsilon Neighborhood ( Ɛ ) and minimum number
of lines (MinLns). The main advantage of this algorithm is that it can predict future
movement of moving objects. This means it can answer questions like, “Where will
this object end up being?” On the contrary, this algorithm cannot give indication
about the temporal factor meaning it is time relaxed.