Keywords

environmental data mining and assimilation, geostatistics, machine learning

Start Date

1-7-2002 12:00 AM

Abstract

The paper presents some contemporary approaches to the spatial environmental data analysis, processing and presentation. The main topics are concentrated on the decision–oriented problems of environmental and pollution spatial data mining and modelling: valorisation and representativity of data with the help of exploratory data analysis, topological, statistical and fractal measures of monitoring networks, spatial predictions and classifications, probabilistic and risk mapping, development and application of conditional stochastic simulation models. The set of tools used consists of machine learning algorithms (MLA) – Multilayer Perceptron, General Regression Neural Networks, Probabilistic Neural Networks, Radial Basis Function Networks, Support Vector Machines and Support Vector Regression, and recently developed geostatistical predictive and simulation models. The innovative part of the report deals with integrated/hybrid models, including ML Residuals Kriging/Cokriging predictions, ML Residuals Simulated Annealing/Sequential Gaussian simulations. The objective of the integrated models is twofold: from one side ML algorithms efficiently solve problems of spatial non-stationarity, which are difficult for geostatistical approach; from another side geostatistical tools are widely and successfully applied to characterise the performance of the ML algorithms, analysing the quality and quantity of the spatially structured information extracted from data by ML. Moreover, mixture of ML data driven and geostatistical model based approaches are attractive for decision-making process.

COinS
 
Jul 1st, 12:00 AM

Environmental Data Mining and Modelling Based on Machine Learning Algorithms and Geostatistics

The paper presents some contemporary approaches to the spatial environmental data analysis, processing and presentation. The main topics are concentrated on the decision–oriented problems of environmental and pollution spatial data mining and modelling: valorisation and representativity of data with the help of exploratory data analysis, topological, statistical and fractal measures of monitoring networks, spatial predictions and classifications, probabilistic and risk mapping, development and application of conditional stochastic simulation models. The set of tools used consists of machine learning algorithms (MLA) – Multilayer Perceptron, General Regression Neural Networks, Probabilistic Neural Networks, Radial Basis Function Networks, Support Vector Machines and Support Vector Regression, and recently developed geostatistical predictive and simulation models. The innovative part of the report deals with integrated/hybrid models, including ML Residuals Kriging/Cokriging predictions, ML Residuals Simulated Annealing/Sequential Gaussian simulations. The objective of the integrated models is twofold: from one side ML algorithms efficiently solve problems of spatial non-stationarity, which are difficult for geostatistical approach; from another side geostatistical tools are widely and successfully applied to characterise the performance of the ML algorithms, analysing the quality and quantity of the spatially structured information extracted from data by ML. Moreover, mixture of ML data driven and geostatistical model based approaches are attractive for decision-making process.