Efficient Spatial-Temporal Analysis of Environment and Public Health Related Data

与环境和公共健康相关的数据属于同一类时空数据，除了包含描述对象特征的常规属性信息外，还包含关于对象的地理位置和时间特征的空间信息。这项拟议工作的重点是分析来自环境和公共健康相关数据的信息，有效的技术对于基于大型时空数据集进行决策的组织至关重要。有效模型的应用在环境保护和公众健康方面都是有用的。传统的数据分析和数据挖掘技术没有对空间背景和时间效应进行建模，可能会导致在空间和时间上系统变化的残差。得出的模型可能不仅是有偏见和不一致的，而且可能与数据集不太匹配。传统的解决空间数据分析的方法是使用经典的数据分析工具，将空间滞后或误差作为解释变量之一。这些技术最大限度地提高了分类精度，但空间精度可能更重要。在大多数这些预测模型中，时间和属性精度被忽略。提出了一种新的计算效率高的时空预测框架ST-POUMS，该框架将地图相似度(包括空间相似度、时间相似度和属性相似度)最大化，而不是分类/预测的准确率。这项工作解决了如何将时空自相关这一时空数据的特征属性结合到ST-PLUMS框架中。ST-PLUMS框架使用一种新的地图相似性度量来搜索模型的参数空间，该度量更适合于时空数据。除了对空间精度进行建模外，ST-PLUMS还可以扩展为在模型中包含时间和属性精度。ST-POMS将为分析时空数据时遇到的困难提供解决方案。它能够处理多维(即空间、属性、时间等)。环境与公共卫生相关数据具有复杂的数据结构，并在海量数据的情况下实现高效率。环境与公共健康相关数据的高效时空分析新技术具有深远的意义。建议的框架由基础理论研究和严谨的实证研究组成，以验证所有概念。这些实验将由一系列日益复杂的案例研究推动，包括禽流感调查的栖息地估计和哮喘入院预测。这些解决方案将对环境保护、犯罪学和司法、房地产管理和环境流行病学等几个重要领域产生直接影响。提出了分析多维、自相关、大容量时空数据的概念、设计、算法和策略。拟议的研究还可能对其他自然科学和社会科学产生广泛影响，这些科学可以利用各种时空数据提供的优势。

ABSTRACTNSF-0513669Wu, WeiliEnvironment and public health related data belong to the same category of spatial-temporal data, which contain spatial information about the geographic location and temporal feature of an object in addition to the conventional attribute information describing the object's characteristics. Efficient techniques for analyzing information from the environment and public health related data, the focus of this proposed work, are crucial to organizations, which make decisions based on large spatial-temporal data sets. The applications of efficient models can be found useful in environment conservation and public health. Traditional data analysis and data mining techniques, which do not model spatial context and temporal effect, may lead to residual errors that vary systematically over space and time. The models derived may turn out to be not only biased and inconsistent, but may also be a poor fit to the data set. The traditional approaches towards solving spatial data analysis are to use classical data analysis tools by using spatial lag or error as one of explanatory variables. These techniques maximize classification accuracy, but spatial accuracy may be of more importance. Temporal and attribute accuracies are ignored in most of these predictive models. In addition, these approaches are often computationally expensive and are confounded with a large datasets.A new computationally efficient spatial-temporal framework, ST-PUMS (Spatial-temporalPrediction Using Map Similarity), which maximize map similarity (including spatial similarity, temporal similarity, and attribute similarity) instead of classification/prediction accuracy was proposed. This work addresses how spatial-temporal autocorrelation, the characteristic property of spatial-temporal data, can be incorporated in the ST-PUMS framework. ST-PUMS framework searches the parameter space of models using a new map-similarity measure that is more appropriate in the context of spatial-temporal data. In addition to modeling spatial accuracy, ST-PUMS can also be extended to incorporate temporal and attribute accuracies in the model. ST-PUMS will provide a solution for the difficulties encountered in analyzing spatial-temporaldata. It is able to cope with multidimensional (i.e., spatial, attribute, temporal, etc.) environment and public health related data with complex data structure, and to achieve high efficiency with large volumes of data.New techniques of the efficient spatial-temporal analysis of environment and public health related data are profound. The proposed framework consists of basic theoretical research as well as rigorous empirical studies to validate all the concepts. The experiments will be driven by a series of increasingly sophisticated case studies, including habitat estimation for bird flu investigation, and asthma hospital admission predictions. The solutions will have a direct impact on several important areas, such as environmental conservation, criminology and justice, real estate management and environmental epidemiology. The concepts, designs, algorithms and strategies are devised to analyze the multiple-dimensional, auto-correlative, large-size spatial-temporal data. The proposed research may also have broad impacts on other natural and social sciences that could take the advantages offered by varieties of spatial-temporal data.