Visual mining of moving flock patterns in large spatio-temporal data sets using a frequent pattern approach

Abstract

The popularity of tracking devices continues to contribute to increasing volumes of spatio-temporal data about moving objects. Current approaches in analysing these data are unable to capture collective behaviour and correlations among moving objects. An example of these types of patterns is moving flocks. This article develops an improved algorithm for mining such patterns following a frequent pattern discovery approach, a well-known task in traditional data mining. It uses transaction-based data representation of trajectories to generate a database that facilitates the application of scalable and efficient frequent pattern mining algorithms. Results were compared with an existing method (Basic Flock Evaluation or BFE) and are demonstrated for both synthetic and real data sets with a large number of trajectories. The results illustrate a significant performance increase. Furthermore, the improved algorithm has been embedded into a visual environment that allows manipulation of input parameters and interactive recomputation of the resulting flocks. To illustrate the visual environment a data set containing 30 years of tropical cyclone tracks with 6 hourly observations is used. The example illustrates how the visual environment facilitates exploration and verification of flocks by changing the input parameters and instantly showing the spatio-temporal distribution of the resulting flocks in the Space-Time Cube and interactively selecting, querying and saving the resulting flocks for further analysis and verification.

Comprehensive Study of Applications of Spatio Temporal Data Mining in Gis

“…Themethodology uses neighbourhood rule These along with relationships are used to create decision tables.The temperature is used as a parameter [8] UlanbekTurdukulov et al have used the pattern generated moving flocks as frequent pattern. Their comparison with other real data sets, the results were far better than previous techniques [9].…”

Enumeration of complete set of flock patterns in trajectories

“…Although the BFE algorithm seems to achieve polynomial time per patterns as ours, it requires exponential space due to table look-up in order to avoid duplicated answers. Turdukulova, Romero, Huismanc, and Retsiosa [12] study the efficient implementation of complete mining of flock patterns based on frequent itemset mining algorithm, such as LCM [13].…”

“…Since the seminal work of Laube et al [8] on a variation of group patterns with duration length k = 1, called flock, meet, diverge, and leadership, there have been a number of efficient algorithms for mining flock patterns in an input collection of 2-dimensional trajectories [3,7,12,14]. However, there seem no algorithms that simultaneously achive complete mining and polynomial delay and space complexity for subclasses r-flock patterns for arbitrary dimension d ≥ 2 and duration length k ≥ 1.…”

Effective Following Patterns Mining Scheme for the Movements of Objects

“…While the method has been focused on mining following patterns between two moving objects, would not be practical in real life. Ulanbek et al [58] used the FPM approach for discovering moving flock patterns in large spatiotemporal data sets. The FPM approach showed to be useful to deal with the problems found in the BFE (Basic Flock Evaluation) algorithm for data sets with a large numbers of trajectories.…”