Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Computing | en_US |
| dc.contributor.advisor | Chan, C. C. Keith (COMP) | - |
| dc.creator | Wu, Pak Kit | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/9935 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | - |
| dc.rights | All rights reserved | en_US |
| dc.title | Discovering spatio-temporal patterns in multivariate spatial time series | en_US |
| dcterms.abstract | A multivariate spatial time series (MSTS) consists of a collection of values by a set of geographical coordinates accompanied by a set of multivariate time series (MTS). An MTS is composed of a number of temporally interrelated variables monitored over a period of time at successive time instants spaced at uniform time intervals. MTS data are generated massively due to recent developments in sensor and satellite technologies, medical measurements, climate informatics, and bioinformatics. These large-scale data encode important information about complex relations among individual time series. Many of these MTS are spatiotemporal by nature in which they are collected together with spatial location information such as latitude and longitude. For example, climate data are from sensors located in different regions, each of which collects periodic readings of variables such as humidity, wind speed, temperature, and rainfall intensity. A computational technique that is able to discover interesting patterns in MSTS data can lead to many applications in diverse areas of research and be helpful to society as well as to the economy. MSTS can be represented as a set of MTSs each of which is associated with a spatial location. Conventional time series analysis methods which consider only the time domain are often adopted to analyze MTS, but the spatial and temporal relationships associated with the individual time series in MSTS are usually ignored, or treated separately, during the pattern discovery process. For this reason, new effective techniques are required. In this thesis, we proposed some such techniques, in particular, that can be used to address the problems of identifying interesting patterns in MSTS and the classification and clustering of them. One of the classical examples of MTS is spatial trajectory data with x coordinate and y coordinate forming the different components of the MTS. In many cases, such data is also spatio-temporal as it may be associated with many spatiotemporal parameters such as velocity and direction etc. Mining spatial trajectories can have many applications in a variety of research areas. For example, in traffic data, finding patterns of driving behavior of moving objects can provide insight into many real applications such as auto insurance and vehicle safety checks. In this regard, we propose in this thesis a technique that can discover association patterns from the feature space characterizing the spatial trajectories. These discovered association patterns, treated as the driving behavior on the road, could be used for the classification of drivers. A classification algorithm has been developed based on the proposed technique to consider the variable length of multiple spatial trajectories of each driver to determine the class membership that exists between association patterns of these trajectories and the class. Integrated with an information theoretic measure, this classifier is able to predict and identify uniquely a driver based on driving patterns of unlabeled trajectories that are for or against a certain class membership. For performance evaluation, we have used it to solve problems in driver classification using their spatial trajectory data. | en_US |
| dcterms.abstract | According to empirical studies on spatial data analysis, mining of MSTS should consider the spatial nature of the objects to be analyzed, their characteristics of the feature space and the uncertainty between the spatial units and their complex features. The proposed technique, to discover association patterns in MSTS, should consider both spatial and temporal information. This proposed technique not only can uncover the temporal and spatial association relationships of MSTS but also can tackle supervised and unsupervised learning tasks. This technique incorporates an initial MTS pattern mining algorithm to detect temporal association relationships from frequent patterns in a set of MTSs for each location. We have developed an algorithm to detect co-occurrence of the discovered temporal patterns across locations by mining a transformed spatiotemporal pattern matrix (STPM) that characterizes the feature space to form spatio-temporal patterns. That is to say, if the frequency of co-occurrence of the respective temporal patterns in different spatial units is significantly higher, the co-occurrence of the temporal patterns across locations is the spatial association patterns of interest. To determine if the frequency of their co-occurrences is significantly higher, we apply a statistical significance test to measure how significantly the observed frequency of the co-occurrences deviates from its expected frequency. Furthermore, we effectively integrate this spatio-temporal pattern-mining algorithm for classification and clustering by an information theoretic measure. If the set of MSTS is labeled, the discovered patterns can be weighted to support or against a certain class membership for the construction of a classifier. If the set of MSTS is unlabeled, the discovered patterns in one location are compared against those discovered in the others so that, by taking the spatial contiguity between locations into consideration, MSTS that have similar discovered patterns and are closer to each other are grouped together into the same cluster. To evaluate the performance of the algorithms, we have tested them on both synthetic and real-world data sets. We have also applied them to tackle several practical problems in some case studies. Both experimental results and findings from practical case studies show the proposed techniques to be promising for MSTS analysis. | en_US |
| dcterms.extent | xxi, 186 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2019 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.LCSH | Multivariate analysis | en_US |
| dcterms.LCSH | Time-series analysis | en_US |
| dcterms.LCSH | Data mining | en_US |
| dcterms.accessRights | open access | en_US |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 991022208058203411.pdf | For All Users | 5.67 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/9935