Human behavior prediction is a rapidly evolving field with numerous applications in various domains. The rapid development of machine learning facilitates research on data-driven behavior prediction. At the same time, in human factors domains, researchers still focus on modeling psychological decision-making processes from the psychological perspectives of view based on subjective data. However, it is difficult to obtain one’s psychological states in real-world scenarios, given that the collection of subjective data in real-world settings is difficult and intrusive. Thus, psychology theories can hardly be applied to behavior prediction directly. To bridge the gap between data-driven behavior prediction and psychological decision-making models, we proposed a novel Psychology-powered Explainable Neural network (PEN). In PEN, for the first time, the psychological factors of individuals (i.e., human-side features such as attitudes to technologies) are modeled and recovered explicitly based on one’s historical behaviors in certain scenarios. The multi-task optimization ensures each component in PEN to be optimized with proper gradient. Assessment results over three evaluation protocols demonstrate that, based on the historical behaviors of an individual only, the PEN can outperform the best existing behavior prediction models by 1.87% and 4.68% on two datasets when a 1:9 training–testing data partition was adopted. The framework in PEN has provided a new solution for explainable human behavior prediction when psychological information is not directly available.