This paper presents a novel human gaze prediction algorithm based on computer vision and algebraic techniques. The proposed method measures the relative position of the iris within the eye in an image and, by integrating this information with an existing 3D head pose estimation model, estimates the gaze direction. Experiments are conducted in a controlled scenario where different points are displayed on a screen (with known screen coordinates), and participants are instructed to sequentially look at them using different approaches while images capturing their faces are taken. By extracting the head and eye pose descriptors from these images and combining them with the known screen coordinates, a gaze prediction model is trained. After the validation process, as a result, the introduction of the new eye pose descriptors in the model increases its accuracy and precision by 10% and 28%, respectively, making the system more robust. This research is part of the Spanish-funded project “DivInTech” and its future aim is to adapt the developed model for children with autism (ASD). Such model will subsequently be used in this project to evaluate the interaction between the study subjects (children with ASD) and the humanoid robot NAO during different educational activities.