Scholarlink

In LoRaWAN networks, shared access channels can lead to packet collisions, impacting overall network performance. This paper presents SmartLoRaML, a technique utilizing machine learning (Random Forest and Gradient Boosting with Stratified cross-validation) for optimized spreading factor (SF) allocation, which is essential while considering packet collisions in LoRaWAN. The technique is implemented using a dataset generated from a LoRaWAN simulator. The simulator code is then modified to incorporate SmartLoRaML. Performance is assessed across various scenarios using accuracy, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), Packet Delivery Ratio (PDR), and energy consumption. Evaluations show that performance varies based on node density and communication radius. For instance, at a 3000 m radius with 100 nodes, Gradient Boosting achieved precision (0.951) and recall (0.88) with an accuracy of 97.8%, while Random Forest demonstrated high accuracy (97.3%). Increasing the radius (e.g., 10 km) reduced classifier effectiveness due to network density. Packet Delivery Ratio (PDR) results showed SmartLoRaML achieving PDRs from 97.8% to 71.9%, influenced by node density and network radius. Also, SmartLoRaML has enhanced the transmit energy consumption compared to a random SF Allocation method used by the simulator. This work provides insights for optimizing LoRaWAN deployments and enhancing IoT network anomaly detection.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Learning to Fly with SmartLoRaML: A Machine Learning Solution for LoRaWAN Spreading Factor Allocation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Usama Mustafa,
  • Sana Mustafa,
  • Bushra Tariq,
  • Imran Rashid

摘要

In LoRaWAN networks, shared access channels can lead to packet collisions, impacting overall network performance. This paper presents SmartLoRaML, a technique utilizing machine learning (Random Forest and Gradient Boosting with Stratified cross-validation) for optimized spreading factor (SF) allocation, which is essential while considering packet collisions in LoRaWAN. The technique is implemented using a dataset generated from a LoRaWAN simulator. The simulator code is then modified to incorporate SmartLoRaML. Performance is assessed across various scenarios using accuracy, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), Packet Delivery Ratio (PDR), and energy consumption. Evaluations show that performance varies based on node density and communication radius. For instance, at a 3000 m radius with 100 nodes, Gradient Boosting achieved precision (0.951) and recall (0.88) with an accuracy of 97.8%, while Random Forest demonstrated high accuracy (97.3%). Increasing the radius (e.g., 10 km) reduced classifier effectiveness due to network density. Packet Delivery Ratio (PDR) results showed SmartLoRaML achieving PDRs from 97.8% to 71.9%, influenced by node density and network radius. Also, SmartLoRaML has enhanced the transmit energy consumption compared to a random SF Allocation method used by the simulator. This work provides insights for optimizing LoRaWAN deployments and enhancing IoT network anomaly detection.