In Nuclear Magnetic Resonance imaging, eddy current is induced in the surrounding conductors when the time varing gradient field is applied. This eddy current generates undesirable secondary gradient fields which often result in temporal as well as spatial inhomogeneity. The induced eddy currents decay exponentially with multiple tiem constants depending on the geometry of surrounding conductors such as the main magnet and the coil configurations. In this thesis, the Linear Prediction Singular Value Decomposition algorithm is applied for the analysis of eddy currents. The time constants and related gains which would be useful for the adjustment of the eddy current compensator are extracted by using this algorithm. This numerical technique enables us to correct or compensate the eddy current efficiently. Furthermore, more precise correction was possible by use of this method than the conventional technique such as the iterative correction. Experimental scheme and its results obtained with KAIS 2.0 Tesla whole body NMR imaging system are also presented.