A first electronic device according to various embodiments may select one of a plurality of representation matrices and one of a plurality of measurement matrices on the basis of a pattern and/or feature of data received from a sensor. The selection of the representation matrix and the measurement matrix may be performed on the basis of machine learning. Based on the selected representation matrix and measurement matrix, the first electronic device may adaptively compress at least a portion of the data. A second electronic device according to various embodiments may restore compressed data on the basis of the result of selecting the representation matrix and the measurement matrix. By dynamically selecting the representation matrix and the measurement matrix on the basis of machine learning, it is possible to reduce an error in the data restored by the second electronic device (e.g., a restoration error).