A conventional way to find a proper grasp to grab and hold any object is to measure its stability which usually is based on physical constraint called force-closure. This execution works well from the theoretical point of view but often fails on an actual robot due to many reasons such as intrinsic errors in robot?s system and a disparity between real and simulated physics. Several research works introduced methods to alleviate those issues and increase the success rate of grasping for a real robot. Caging and Independent Contact Region are ones of them. In this work, we investigate a method to find a grasp that is more stable and robust by combining those two techniques which result in a learning-based approach that utilizes an artificial neural network to find a proper grasp based on those techniques and some heuristic methods.