ИСТИНА

One of the main problems in exploration geophysics is restoration of distribution of some physical value inside the thickness of the earth by the values of physical fields measured on its surface. Such inverse problems are well solved with machine learning methods. However, such methods are often sensitive to the number of input features; so, reducing the input dimensionality is required. In this study, we compare use of several feature selection methods when solving an inverse problem of magnetotelluric sounding, and make some conclusions on their proper use.