Context-Aware Machine Learning Framework for ADR Optimization in LoRaWAN IoT Networks

LoRaWAN (Long Range Wide Area Network) faces scalability and reliability challenges in heterogeneous Internet of Things (IoT) deployments due to interference, device mobility, and diverse traffic demands. Existing Adaptive Data Rate (ADR) schemes such as Standard Adaptive Data Rate (Standard-ADR) and Context-Aware Adaptive Data Rate (CA-ADR) optimize link parameters but remain traffic-agnostic or location-dependent. This paper presents a clustering-assisted, three-stage classification framework for class-aware ADR that relies solely on the Received Signal Strength Indicator (RSSI) and Signal-to-Noise Ratio (SNR), requiring no GPS metadata. Devices are clustered, refined via Shapiro–Wilk quartile analysis, and classified using a Support Vector Machine (SVM). The resulting labels drive a policy selector integrated with multiple ADR backends: Standard-ADR, CA-ADR, and two new variants, Coati Optimization Algorithm–based ADR (CoATI-ADR) and Binary Coati Optimization Algorithm–based ADR (Binary CoATI-ADR), derived from the Coati Optimization Algorithm (COA) and its binary form (BCOA) for discrete adaptation of spreading factor and transmit power. Simulations across industrial, suburban, and rural–agricultural scenarios show improvements of up to 2% in Macro-Averaged F1, 3–5% in Packet Delivery Ratio (PDR), and 15–25% in energy efficiency, sustaining fairness near 1.0. The framework thus provides a scalable, location-free, and traffic-aware ADR strategy suited to large-scale IoT networks.