<p>Inverse kinematics calculations are of great importance to control robotic systems. In more specific applications such as vibration control and calculating endpoint forces, online inverse kinematics calculations are required, instantaneously. Geometric, iterative, and algebraic approaches might be insufficient to provide the desired speed or performance in complex robot structures such as a 6-degree-of-freedom industrial robot. Artificial neural networks (ANNs) are an appropriate alternative for pose prediction and trajectory generation to reduce computational cost. In the present study, cascade networks architecture is introduced. The novelty of this study is that the proposed architecture aims to reduce positioning errors by estimating each joint angle sequentially starting from the base considering the kinematic chain of the robot. The performance of ANNs for inverse kinematics prediction on precision was investigated. The proposed ANN architecture is compared with existing approaches. Neural networks are trained using the dataset obtained by forward kinematics calculations using the kinematic model of the Yaskawa GA50 industrial robot. Hyperparameters were systematically optimized through multiple observations. The optimized models to predict the solutions in joint space were evaluated for two different robotic trajectories. The proposed cascade neural network architecture improves the precision of the solutions.</p>