Synthetic populations are instrumental in analyzing natural populations, with agent-based microsimulations being their most prevalent application. These simulations serve various fields, including transit planning, epidemiology, and urban development. Traditionally, population synthesis methods have been plagued by the generation of overly similar populations and high computational demands. Recent advancements in machine learning offer promising alternatives, enabling the creation of diverse agents that maintain statistical significance while considerably reducing computation time, specially for populations with numerous attributes. A notable advancement in this domain is the integration of temporal dynamics into population synthesis, leading to the concept of Dynamic Synthetic Population. Despite the longstanding exploration of demographic update methods, their seamless integration with population synthesis remains a challenging and open question. The incorporation of machine learning techniques holds the potential for addressing this gap. This research proposes a novel methodology for developing a Dynamic Synthetic Population Framework specifically tailored to the Metropolitan Area of Porto. The proposed framework uses machine learning to generate temporally dynamic synthetic populations with reliable activity chains. This dynamic synthesis approach facilitates comprehensive analyses of population changes. It supports the exploration of various future and past scenarios, thereby enhancing the utility and accuracy of microsimulations in diverse fields.