Low-power wide-area networks (LPWANs) constitute a variety of modern-day Internet of Things (IoT) applications. Long range (LoRa) is a promising LPWAN technology with its long-range and low-power benefits. Performance enhancement of LoRa networks is one of the crucial challenges to meet application requirements, and it primarily depends on the optimal selection of transmission parameters. Reinforcement learning-based multi-armed bandit (MAB) is a prominent approach for optimizing the LoRa parameters and network performance. In this work, we propose a new discounted upper confidence bound (DUCB) MAB to maximize energy efficiency and improve the overall performance of the LoRa network. We designed novel discount and exploration bonus functions to maximize the policy rewards to increase the number of successful transmissions. The results show that the proposed discount and exploration functions give better mean rewards irrespective of the number of trials, which has significant importance for LoRa networks. The designed policy outperforms other policies reported in the literature and has a lesser time complexity, a comparable mean rewards, and improves the mean rewards by a minimum of 8%. © 2023 by the authors.