In this paper a novel adaptive bandwidth allocation scheme called Prudent Creditization Polling (PCP) is proposed and adapted with the reservation protocol IPACT. PCP attempts to reduce the average queue length at the ONUs and the average packet delay throughout the network. At each polling period, PCP attempts to creditize the request of the ONUs that suffer from high RTT, high average delay per packet, and high buffer growth. We implemented PCP with a fuzzy functional approach and a linear regression approach. We simulated PCP and several service disciplines in IPACT and compared their performance in terms of average packet delay and queue length. We show that PCP improves the performance of IPACT significantly in terms of the performance metrics aforementioned.