A computer-implemented method is presented for arranging a plurality of sensors in a three-dimensional space. The method includes obtaining actual measured values from the plurality of sensors placed in the three-dimensional space, applying group regularization to create one or more models providing consistency between the actual measured values and corresponding predicted values, selecting a model from the one or more models having a given number of groups, wherein each group has significant coefficients, and, according to the selected model, arranging a subset of the plurality of sensors at locations corresponding to the significant coefficients to predict one or more parameters in the three-dimensional space.