Comparison of four machine learning algorithms for spatial data analysis

This chapter proposes a clear methodology on how to use machine learning algorithms for spatial data analysis in order to avoid any bias and eventually obtain fair estimation of their performance on new data. Four different machine learning algorithms are presented, namely multilayer perceptrons (MLP), mixture of experts (ME), support vector regression (SVR) and a local version of the latter (local SVR). Evaluation criteria adapted to geostatistical problems are also presented in order to compare adequately different models on the same dataset. Finally, an experimental comparison is given on the SIC97 dataset as well as an analysis of the results.