In this research, we constructed an inverse design framework to determine the optimal composition of steel for specific creep properties. The data, sourced from existing literature, comprised a broad array of chemical compositions and mechanical properties, encompassing 291 instances with 17 independent variables and one dependent variable. Utilising this comprehensive dataset, we built several predictive models through a variety of machine learning algorithms. To discern the intricate relationships amongst the variables, we performed Pearson correlation coefficient analysis and assessed feature importance. The most accurate prediction model was selected using a grid search approach. With the predictive model established, we applied genetic algorithms to inversely deduce the steel composition, informed by the predictive model's insights. This methodological fusion presents a novel pathway to material design, potentially expediting the development of new steel grades with the desired properties.