Spatial-temporal data modelling and processing for personalised decision support
Introduction: Capturing the nature of spatio/spectro-temporal data (SSTD) is not an easy task nor is understanding the relationships between the different data dimensions such as between temporal and spatial, temporal and static, and between temporal variables themselves. In the past it has been nor...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2015
|
| Subjects: | |
| Online Access: | http://eprints.uthm.edu.my/9222/ http://eprints.uthm.edu.my/9222/1/Muhaini_Othman.pdf |
| Summary: | Introduction: Capturing the nature of spatio/spectro-temporal data (SSTD) is not an easy task
nor is understanding the relationships between the different data dimensions such as
between temporal and spatial, temporal and static, and between temporal variables
themselves. In the past it has been normal to separate the SSTD dimensions and
only take one dimension of the data and convert it into a static representation and
model from there. While other dimensions are either ignored or modelled separately.
Although this practice has had significant outcomes, the relationships between data
dimensions and the meaning of that relationship defined be the data is lost and can
result in inaccurate solutions. Any relationship between the static and dynamic or
temporal data has been under analysed, if analysed at all, dependent upon the field
of study.
Purpose of the research: The purpose of this research is to undertake the modelling of dynamic data without
losing any of the temporal relationships, and to be able to predict likelihood of
outcome as far in advance of actual occurrence as possible. To this end a novel computational
architecture for personalised (individualised) modelling of SSTD based on
spiking neural network methods (PMeSNNr), with a three dimensional visualisation
of relationships between variables is proposed. The main architecture consists of a
spike time encoding module; a recurrent or evolving 3D spiking neural network reservoir
(eSNNr); an output module for either classification or prediction based around
another evolving spiking neural network; and a parameter optimisation module. In
brief, the architecture is able to transfer spatio-temporal data patterns from a multidimensional
input stream into internal patterns in the eSNNr. These patterns are
then analysed to produce a personalised model for either classification or prediction
dependent on the specific needs of the situation. |
|---|