Application of K-means Algorithm to Two-dimensional Electrical Resistivity Imaging

This paper describes the application of unsupervised classification using k-means algorithm to combine and classify two-dimensional (2D) inverse resistivity models arising from the use three different electrode arrays for resistivity imaging. The 2D inversion results obtained were used as the input images for the classification scheme. The k-means algorithm classified the synthetic models each into sixteen (16) clusters, each cluster associated with three mean resistivity vectors from each electrode array. The resistivity values were assigned to each cluster employing basic statistical approach. An evaluation of the classified images was performed by comparison with the resistivity values between classified and true models. This was further confirmed by estimation of the mean absolute error for each image. The results of comparison of resistivities and the estimated error show that maximum approach give the best representative of the real models than using individual apparent resistivity. Keywords: Unsupervised classification, inverse resistivity model, k-means, clusters, accuracy assessment, classified image