Wang, J; Cheng, T; Haworth, J; (2012) Space-Time Kernels. In: Shi, W and Goodchild, M and Lees, B and Leung, Y, (eds.) Advances in Geo-Spatial Information Science. (25 - 32). CRC Press Taylor & Francis Group: UK.
Full text not available from this repository.
Kernel methods are a class of algorithms for pattern recognition. They play an important role in the current research area of spatial and temporal analysis since they are theoretically well-founded methods that show good performance in practice. Over the years, kernel methods have been applied to various fields including machine learning, statistical analysis, imaging processing, text categorization, handwriting recognition and many others. More recently, kernel-based methods have been introduced to spatial analysis and temporal analysis. However, how to define kernels for space-time analysis is still not clear. In the paper, the relevant kernels for spatial and temporal analysis are first reviewed, and then a space-time kernel function (STK) is presented based on the principle of convolution kernel for space-time analysis. The proposed space-time kernel function (STK) is then applied to model space-time series using support vec-tor regression algorithm. A case study is presented in which STK is used to predict China’s an-nual average temperature. Experimental results reveal that the space-time kernel is an effective method for space-time analysis and modeling.
|Keywords:||Space-Time Kernels, Space-Time Analysis, Support Vector Regression|
|UCL classification:||UCL > School of BEAMS > Faculty of the Built Environment > Centre for Advanced Spatial Analysis|
Archive Staff Only: edit this record