Kinetic predictions for complex reaction systems of hydrocarbons are theoretically and technologically crucial to the petrochemical industry. Among several proposed kinetic models, a lumping kinetic model is a comparatively simple and developed method wherein a complex system is lumped into several pseudo-components. To acquire more accurate mechanistic information, kinetic models at the mechanistic level are developed, such as single-event kinetic and structure-oriented models. However, the number of kinetic parameters increases exponentially in these methods. Lumping kinetic methods are then reexamined, and kinetic models, such as relumping single-event kinetic methods, bimolecular methods, and special pseudo-component methods, are proposed to simplify the reaction system. Many mathematical methods, such as annealing algorithm or artificial neural networks, have also been developed to solve these complex reaction problems. Although a number of complex intrinsic reaction studies have been introduced, the combination of excellent prediction performances and practical industrial applicability remains a central challenge facing this field. This situation motivated this study, to review the recent development of reaction prediction models and their application in industrial processes. Furthermore, the practical applications of these possible pathways of kinetic predictions for mechanistic studies are addressed.