Efeito do tamanho da mostra de treinamento da predição de perfis faltantes por técnica de análise multivariada.

Abstract

Geophysical well logs constitute an important data source for the evaluation of the potential of a given area for underground natural resources exploitation purposes. For several reasons, hardly the complete log set is available, so that there may be missing logs in a given well. However, a geophysical well logs characteristic is that there is some level of redundancy between them, so this feature can be used to estimate a given missing log from other available logs. This work estimate missing logs through the application of the multivariate statistical technique KNN (K-thnearest neighbor) which is based on the measurement of similarity between the values of several logs of a training sample set. The estimated value of a given missing log is obtained by the similarity between the available logs and the training set values. In particular this work evaluates the effect of the training set size on the efficiency of KNN prediction. The results obtained show that increasing the training set size leads to a reduction of the difference between the actual log and the estimated one. However, training set with only 10% of the available data already provide acceptable prediction for logs whose variation occurs in linear scale, as density, sonic, gamma ray and neutron porosity logs. For electrical resistivity log KNN not achieved acceptable results, because this log does not have redundancy with other available logs