Movement Mining - Computational Movement Analysis - page 39

Andrienko and Andrienko ( 2007 ) stipulate the combination of individual movement
behaviors (IMB) to form dynamic collective behaviors (DCB). Irrespective of the
precise nature of the building blocks, there seems to be an agreement that higher-level
behavior patterns emerge when lower-level building blocks are strung together in a
temporally ordered sequence. Merki and Laube ( P16 . 2012 ) even showed that not
only the sequence matters, but that the temporal spacing between lined-up building
blocks plays a crucial role when separating related yet different behaviors.

3.2.3.3 Non-deterministic Pattern Mining

Even though movement pattern mining qualifies as a “retrieval by content” data
mining task (for example “detect all leaders leading m followers for k time units”),
its true strength lies in exploratory rather than confirmatory analysis, in hypothesis
forming rather than hypothesis testing.Movement patterns are a good example of how
movement mining is strongest when prompting researchers to new and unexpected
relationships in data. The exact extent of a leadership pattern is not very relevant.
The knowledge that in a certain data set there exist leadership patterns in the first
place, and the order of magnitude of such patterns, is much more relevant.
Similarly, in most cases there is not apriori knowledge about the precise extent
or any parameter or threshold specifying a movement pattern. How many followers
must an alpha wolf have and for how long must they follow it? In some cases the
domain specific literature may give indications about parameters specifying move-
ment patterns. However, it is the repeatability of the algorithmic search for such
patterns that allows for series of sensitivity experiments, lowering the influence of
potentially arbitrarily chosen thresholds in the knowledge discovery process. Merki
and Laube ( P16 . 2012 ), for example, assessed the sensitivity of the parameters

,the
angle of the front region for pursuit and escape , and delay d between an approach
and a separation in a confrontation pattern. Laube et al. ( P12 . 2011a ) explicitly
studied the sensitivity of the data mining process with respect to the chosen size of
flocking patterns (here with grazing cows). Performing movement mining in such an
exploratory way turns a potential weakness into a strength as it allows exactly the
required embedding of the somewhat mechanistic pattern mining in further domain
expertise, through a close collaboration of data mining experts with domain experts
(Fayyad et al. 1996 , p. 39).

φ

3.2.4 Exploratory Analysis and Visualization

Given its specific characteristics, movement data presents an ideal use case for
spatio-temporal exploratory analysis, visualization, and visual analytics concepts
(Andrienko et al. 2010 ). Irrespective of the precise label, the core idea is to combine
the strengths of human and computational data processing (Keim et al. 2008 ). Just as
is outlined in the previous sections, algorithmic techniques are used to prepare and