ARFOR combines adaptive random forests and owl optimization for energy-efficient WSN-IoT routing

A New AI Routing System Could Keep IoT Sensor Networks Alive 53% Longer

Tiny wireless sensors are becoming the nervous system of modern infrastructure, quietly measuring heart rates, electricity demand, air quality, temperature and countless other variables. Yet the networks that connect these devices face a stubborn practical limit: most sensor nodes run on small batteries that are difficult or impossible to replace. A new routing framework called ARFOR is designed to make those batteries last longer by combining machine learning with a nature-inspired optimization algorithm. In simulations, the system increased residual network energy by more than 33%, cut total energy consumption by 53% and improved the packet delivery ratio by more than 32% compared with recent state-of-the-art protocols. The findings, reported by an international research team in Cluster Computing, could have implications for sensor networks used in healthcare, smart grids and environmental monitoring, where communication failures can be more than an inconvenience.

Wireless sensor networks, or WSNs, consist of distributed nodes that collect measurements and relay them, often through multiple intermediate devices, to a central sink or gateway. Unlike conventional computers or smartphones, these nodes typically operate with limited processing power, memory, bandwidth and battery capacity. Sending data can be particularly expensive because a radio must remain active and amplify signals across a distance; in many network models, transmission energy rises rapidly as the separation between nodes increases. A node that repeatedly serves as a relay can therefore drain its battery much faster than its neighbors. Once several strategically placed nodes fail, the network may become partitioned, leaving healthy sensors unable to reach the gateway. This “hotspot” problem is intensified when nodes move, hardware fails, environmental interference changes wireless links or traffic suddenly surges.

ARFOR addresses the problem as a constantly changing decision process rather than a one-time search for the shortest route. Its name refers to the combination of an enhanced Random Forest model and the Owl Optimization Algorithm, or OOA. Random forests are ensemble machine-learning models that combine the decisions of many decision trees. Each tree examines patterns in input data and produces a prediction, while the forest aggregates those predictions to improve robustness. In ARFOR, the model is augmented with shared memory and an energy-aware attention mechanism. In practical terms, that means the system is intended to retain useful information from earlier network states while assigning greater importance to variables related to energy and changing connectivity. The model predicts likely topology changes and proposes initial routes before the optimization stage refines them.

The topology of a sensor network is the pattern of links connecting its nodes, and it is rarely fixed in real-world IoT deployments. A medical monitoring system may experience interference as a patient moves through a hospital. A smart-grid network may be affected by obstacles, electromagnetic noise or equipment failures. Environmental sensors can be separated by large distances and exposed to weather that alters radio performance. ARFOR is designed to process multiple forms of network information at once, including the condition of available paths and the energy state of participating nodes. By anticipating topology changes rather than reacting only after a link breaks, the Random Forest component can provide the optimizer with a more useful starting point. The goal is to avoid routes that appear efficient in the moment but are likely to become costly or unreliable a short time later.

OOA supplies the second half of the framework. Nature-inspired optimization methods translate behavioral patterns observed in animals or other natural systems into mathematical search strategies. The Owl Search Algorithm, on which OOA is based, is a heuristic approach for global optimization. In an optimization problem, a “solution” can be represented as a candidate route, while the quality of that solution is measured by an objective function. The objective may combine several competing criteria, such as energy consumption, remaining battery power, link reliability and the probability that a packet reaches its destination. A route with fewer hops is not necessarily the best route if it forces one low-energy node to transmit repeatedly. Conversely, a longer route may be preferable if it distributes communication work across healthier nodes. OOA searches among these alternatives and attempts to move toward combinations that offer a better overall balance.

The distinctive feature of ARFOR is the interaction between prediction and search. Rather than allowing the optimization algorithm to explore the entire space of possible routes from scratch, the enhanced forest supplies an informed initial solution based on the network’s recent behavior. OOA then refines that proposal using multimodal network data, with its parameters adjusted dynamically as conditions change. The process forms a closed feedback loop: the network state influences the predicted routes, the optimizer selects improved paths, and new network information is fed back into later decisions. Such a design is important because wireless routing is a multi-objective problem. Maximizing packet delivery alone could require expensive retransmissions or energy-intensive paths, while minimizing energy alone could sacrifice reliability. ARFOR seeks a compromise that remains responsive as the network evolves.

The reported performance gains are striking, but they come from simulations rather than a long-term deployment across physical sensor hardware. The study’s abstract states that ARFOR was compared with recent state-of-the-art protocols and produced more than a 33% increase in residual energy, a 53% reduction in total energy consumption and an improvement of more than 32% in packet delivery ratio. Residual energy refers to the battery power left in the network after operation, making it a direct indicator of how much longer the system might continue functioning. Total energy consumption captures the power spent on communication and related network activity. Packet delivery ratio measures the fraction of transmitted packets that successfully arrive at their intended destination. Improving all three metrics simultaneously is challenging because a protocol can often save energy by reducing transmissions at the cost of lost data. The results suggest that ARFOR’s adaptive route selection avoided enough inefficient or unstable paths to improve both longevity and communication reliability in the tested scenarios.

If the gains translate to real deployments, the technology could be especially valuable where maintenance is expensive or dangerous. In healthcare, wireless body-area or clinical sensor systems may monitor vital signs and other physiological signals, making reliable delivery essential. In smart grids, distributed sensors can help track electricity flows, detect faults and support demand management across broad areas. Environmental monitoring networks may be scattered through forests, agricultural fields, industrial sites or remote ecosystems, where replacing batteries can require substantial labor and travel. A routing system that spreads communication costs more evenly could delay the failure of individual nodes and reduce the risk of network partitioning. The authors specifically identify healthcare and smart-grid systems as promising application areas, although the study does not establish clinical or grid-level performance and does not demonstrate that ARFOR is ready for direct operational use.

The approach also exposes the broader challenge of placing artificial intelligence inside severely constrained devices. Machine-learning models can improve decisions, but computation itself consumes energy, and sensor nodes may not have the memory or processor capacity required to run complex models continuously. ARFOR’s use of an enhanced Random Forest is potentially relevant because tree-based models can be less computationally demanding than some deep-learning systems, but the source material does not provide a full accounting of the framework’s processing overhead, memory requirements or performance on physical hardware. Nor does the reported summary establish how the method performs under malicious traffic, privacy constraints, unpredictable radio conditions or very large networks. These questions will matter before an algorithm designed in simulation can be trusted with patient data or critical infrastructure. The study therefore represents a promising routing strategy, not proof that every IoT network can immediately achieve the reported savings.

The research team includes Mehdi Hosseinzadeh, Aso Darwesh, Amir Masoud Rahmani, Mohammad Mohammadi, Amin Mehranzadeh, Thantrira Porntaveetus and Sang-Woong Lee, with affiliations spanning universities and research centers in Vietnam, Azerbaijan, India, Jordan, Taiwan, Iran, Thailand and South Korea. The paper reports no external funding, and the data supporting its findings are available from the corresponding author upon reasonable request. Its central idea is straightforward but powerful: a sensor network should not treat routing as a static map problem when its batteries, links and traffic patterns are constantly shifting. By pairing a forecasting model with a search algorithm capable of balancing multiple objectives, ARFOR attempts to make every packet carry less of the network’s limited energy burden. Whether that promise survives contact with real hardware and unpredictable environments will determine if owl-inspired optimization becomes a practical tool for keeping the IoT running longer.