Owing to advancements in robotics, researchers have been focusing on integrating humanoid robots into actual environments. Most humanoid robots are equipped with seven-degree-of-freedom (DoF) arms that allow them to be flexible in different scenarios. The controller of a 7-DoF robotic arm must select one solution among the infinite sets of solutions for a given inverse kinematics problem. To date, no suitable approach has been developed for identifying appropriate human-like postures for a robotic arm with an offset wrist configuration. In this paper, we propose a novel algorithm that considers the movement of the human arm to consistently find a suitable human-like posture. First, a one-class support vector machine model is employed to classify human-like postures. Then, the algorithm uses the redundancy characteristic of a 7-DoF robotic arm with a linear regression model to enhance the search of human-like postures. Finally, the proposed algorithm is demonstrated in simulation, where it successfully optimized point-to-point trajectories by modifying only the endpoint posture.