Importance of Features Selection in Application of Pattern Recognition in Estimating the Depth of Anticline Using Gravity Data

Pattern recognition algorithms especially neural network in geophysical interpretations and other Earth sciences have been used since some years ago. In neural network and other pattern recognition algorithms like support vector classifier (SVC) that the latter method is used in this research, by using the values of the features, which has been extracted from the objects (in our work gravity profiles are objects), classification of the objects can be done. Usually the features are selected subjectively. In this paper, we have presented a homemade software that can select proper features objectively. By using SVC and the mentioned features selection (FS) software, depth estimations of anticlines have been done in this research. We have shown the difference of using proper features and improper ones in the mentioned depth estimation (a kind of classification). In this paper, twenty synthetic gravity profiles with anticline shape sources are created for training SVC and the same amountof synthetic profiles are created for testing. It has shown that depth estimation with proper features is more precise than depth estimation with improper features. Also it should be emphasized that FS is important not only in depth estimation of anticlines, but also in all kinds of classifications in Earth sciences and the mentioned homemade software code is applicable in all of them.