Back to top

Maritime Data Analytics


Preparing the Maritime Data for Analytics Purposes


Input to the following sections is a small subset of trajectories in an area of Brest, France. [File format].

Getting Familiar with the Data


An example trajectory with error in the reported location. [File format].
An example trajectory with large temporal and spatial gap. [File format].

Data Cleansing and Transformation


An example trajectory with a systematic error in the reported location. [File format].
An example trajectory with a systematic error in the reported location removed. [File format].
An example of simplification:
The original trajectory before applying simplification. [File format].
The trajectory after the applycation os simplification. [File format].
Applying simplification to the whole current dataset. [File format].


Discovering Knowledge from Maritime Data



Cluster Analysis


Trajectory clusters discovered by T-OPTICS. [File format].
Whole-trajectory clusters discovered by S2T-Clustering. [File format].
Sub-trajectory clusters discovered by S2T-Clustering. [File format].

Group Behaviour


Initial trajectories where the flock discovery algorithm is applied. [File format].
Identified flocks from the applied flock discovery algorithm. [File format].

Hot-spot Analysis


The top-10 hot-spots discovered by THS algorithm for a data set covering the Brest area. The cell size is configured to be 0.01 degrees. [File format].
Visualization of the discovered hot-spots.

Frequent Route Discovery - Network Discovery



Aggregated Method


Trajectories visualized in image. Statistics over trajectories are calculated. [File format].

Then, a Voronoi tesselation is created. Voronoi cells become places visited by the trajectories. [File format].

Next, segments aggregated for each pair of visited places
frequent routes are extracted
also zoomed in. [File format].


Data Mining Method


Trajectories are first transformed from sequences of points to sequences of square cells. Cells belong to a grid with user defined size (5 Km). All points are assigned to the grid cells'

thus trajectories are transformed to sequences of cells. [File format].

A sequential pattern mining algorithm is applied and frequent patterns are discovered. [File format].
Visualization of pattern (1)
Visualization of pattern (2)
Visualization of pattern (3)
Visualization of pattern (4)




Network Discovery


For network discovery, we utilize another dataset which is much richer than the already used. Corresponding image for visualization of input traces. [File format].

Initially, a clustering algorithm is applied to reveal clusters of points. [Clusters file format].
Then these clusters, based on their membership are transformed to network nodes. [File format].
Corresponding image for visualization of network nodes with membership over 500
and with membership over 1500.
Lastly, network edges are discovered by applying a frequent sequential pattern algorithm. [File format].
Corresponding image for visualization of network edges with weight over 500
and with weight over 1500.