Global supply chains suffered considerable impact from the disruptions caused by the pandemic scenario and the need for efficient supply chain management is at an all-time high. Supply chain performance is strongly influenced by the inventory management policy adopted at each echelon. Thus, there is a requirement for an efficient inventory replenishment policy for supply chain environments undergoing frequent changes. The traditional static replenishment policies, omnipresent in industries, prove to be suboptimal in this case as the optimal replenishment policy depends on existing supply chain conditions. This work proposes a dynamic framework, inspired by a similar model identified in literature, which is able to find the optimal replenishment policy for a particular echelon in a three-echelon serial supply chain environment characterised by eight attributes. Supervised machine learning techniques were incorporated for developing the framework which is finally able to select and implement the optimal replenishment policy from four possible alternatives. The dynamic framework is developed using different machine learning algorithms, which further gives an insight into their performance. A comparative analysis of the proposed models is performed under two supply chain environments, namely the chaotic and the fast-changing. The results highlight the improvement obtained by considering the dynamic framework rather than the static alternatives. Additionally, the random forest (RF) based model was found to be the best performing model compared to its counterparts. Finally, the proposed model was observed to outperform the model identified from literature, thus portraying the significance of the work.