Издательство EPFL Press, 2009, -380 pp.The book is devoted to the analysis, modelling and visualisation of spatial environmental data using machine learning algorithms. In a broad sense, machine learning can be considered a subfield of artificial intelligence; the subject is mainly concerned with the development of techniques and algorithms that allow computers to learn from data. In this book, machine learning algorithms are adapted for use with spatial environmental data with the goal of making spatial predictions.
Why machine learning? A brief reply would be that, as modelling tools, most machine learning algorithms are universal, adaptive, nonlinear, robust and efficient. They can find acceptable solutions for the classification, regression, and probability density modelling problems in high-dimensional geo-feature spaces, composed of geographical space and additional relevant spatially referenced features. They are well suited to be implemented as predictive engines for decision-support systems, for the purpose of environmental data mining, including pattern recognition, modelling and predictions, and automatic data mapping. They compete efficiently with geostatistical models in low-dimensional geographical spaces, but they become indispensable in high-dimensional geofeature spaces.
The book is complementary to a previous work [M. Kanevski and M. Maignan, Analysis and Modelling of Spatial Environmental Data, EPFL Press, 288 p., 2004] in which the main topics were related to data analysis using geostatistical predictions and simulations. The present book follows the same presentation: theory, applications, software tools and explicit examples. We hope that this organization will help to better understand the algorithms applied and lead to the adoption of this book for teaching and research in machine learning applications to geo- and environmental sciences. Therefore, an important part of the book is a collection of software tools – the Machine Learning Office – developed over the past ten years. The Machine Learning Office has been used both for teaching and for carrying out fundamental and applied research. We have implemented several machine learning algorithms and models of interest for geo- and environmental sciences into this software: the multilayer perceptron (a workhorse of machine learning); general regression neural networks; probabilistic neural networks; self-organizing maps; support vector machines; Gaussian mixture models; and radial basis-functions networks. Complementary tools useful for exploratory data analysis and visualisation are provided as well. The software has been optimized for user friendliness.Learning from Geospatial Data
Exploratory Spatial Data Analysis. Presentation of Data and Case Studies
Geostatistics
Artificial Neural Networks
Support Vector Machines and Kernel Methods
