A Comprehensive Multiple Linear Regression Modeling and Analysis of LoRa User Device Energy Consumption

The rapid expansion of Long Range (LoRa) and Long Range Wide Area Network (LoRaWAN) protocol technologies in large-scale Internet of Things (IoT) deployments highlights the need for precise and analytically grounded energy consumption (EC) estimation of battery-powered LoRa end devices (DVs). Since LoRa DV instantaneous EC strongly depends on key transmission parameters, primarily including spreading factor (SF), transmit (Tx) power, and LoRa message packet size (PS), accurate modelling of their combined influence is essential for optimizing LoRa end DV lifetime, ensuring energy-efficient network operation, and supporting transmission parameter-adaptive communication strategies. Motivated by these needs, this paper presents a comprehensive multiple linear regression modelling framework for quantifying LoRa end DV EC during one transmission and reception LoRa end DV Class A communication cycle. The study is based on extensive high-resolution electric-current measurements collected over 69 measurement sets spanning different combinations of SFs, Tx power levels, and PS values. Based on measurement results, a total of 14 multiple linear regression models are developed, each capturing the joint impact of two transmission parameters while holding the third fixed. The developed regression models are mathematically formulated using linear, interaction, and polynomial terms to accurately express nonlinear EC behavior. Detailed statistical accuracy assessments demonstrate excellent goodness of fit of the developed EC multiple linear regression models. Complementary numerical analyses of regression models EC data distribution further validate regression models’ reliability, and highlight transmission parameter-driven variability of Lora end DV EC. The results of numerical analyses for LoRa end DV EC data distribution show that specific combinations of SF, Tx power, and PS transmit parameters amplify or mitigate EC differences, demonstrating that their joint variability patterns can significantly alter instantaneous energy demand across operating conditions. These interactions underscore the importance of modelling parameters together, rather than in isolation. The developed regression models provide interpretable mathematical formulations of instantaneous LoRa end DV EC prediction for transmission at different combinations of transmission parameters, and offer practical value for energy-aware configuration, battery-lifetime planning, and optimization of LoRa network-based IoT systems.