Since static lease allocation frequently results in either address exhaustion or underutilization, the need for effective IP address space utilization has increased due to the quick growth of networked devices in large-scale DHCP systems. Conventional approaches use set lease terms that don’t adjust to changing network conditions, which leads to ineffective resource management. Existing heuristic-based strategies, like usage patterns or client-based (spatiotemporal) lease modifications, offer only partial results and are not flexible enough to accommodate varying device behaviors and real-time traffic variations. These techniques frequently require manual tuning, have scalability issues, and are not transferable to different network setups. This research presents the first dynamic lease-time allocation in large DHCP networks that uses machine learning, which is a new and unexplored approach to IP address use. The primary objective is to construct an adaptive ML model that predicts optimal lease periods based on real-time network demand and device behavior. To overcome the drawbacks of rule-based approaches in dynamic network environments, the suggested solution uses predictive analytics to dynamically modify lease times, guaranteeing optimized utilization of the address space without the need for human involvement.