Abstract—This paper proposes a new interpolation method for spatial data based on an adaptive neural networks using only the different of x-coordinate, y-coordinate between observed data and their nearest neighbors, and values of neighbors surrounding unobserved location for training network architecture. Unobserved data are interpolated by function of its absolute location and relative location in x-coordinate and y-coordinate and corresponding value at absolute location of k-nearest neighbors. We compared our new proposed method by using observed data to generate prediction map using simulation data set and real world data set. The experimental results show that, by using relationship between nearest neighbors of unobserved point can achieve the good accuracy compare to competitive method for various data set and at different rate of missing.
Index Terms—Interpolation method, adaptive neural networks, prediction map, geostatistics.
Sathit Prasomphan is with Department of Computer and Information Science Faculty of Applied Science, King Mongkut’s University of Technology North Bangkok, Thailand (e-mail: sathitp@ kmutnb.ac.th).
Shigeru Mase is with Department of Mathematical and Computing Sciences, Tokyo Institute of Technology,Japan(e-mail: email@example.com)
Cite:Sathit Prasomphan and Shigeru Mase, "Generating Prediction Map for Geostatistical Data Based on an Adaptive Neural Network Using only Nearest Neighbors," International Journal of Machine Learning and Computing vol. 3, no. 1, pp. 98-102, 2013.