Excited to share our work "Harnessing Cellular Network Data for Accurate Transportation Mode Recognition" accepted for publication in IEEE Transactions on Vehicular Technology
1 min read
·
About
Excited to share that our work "Harnessing Cellular Network Data for Accurate Transportation Mode Recognition" has been accepted by IEEE Transactions on Vehicular Technology, with co-authors Kalamkas Zhagyparova, Nour Kouzayha, Hesham Elsawy, Ahmed Bader, and Tareq Al-Naffouri.
🔥 In this work, we show that transportation modes — bus, car, walk, and train — can be accurately classified using only radio access network (RAN) measurements from cellular infrastructure, without relying on GPS, IMU sensors, or any user-installed apps.
💡 Key Highlights:
⭐ A privacy-preserving, energy-efficient framework that uses only 4G cellular measurements from serving and neighboring cells, designed for operator-side deployment.
⭐ We introduce CellMob, a large-scale dataset of 160 trips and 55+ hours of travel time across two urban cities (Makkah, Jeddah) and a semi-rural campus (KAUST) in Saudi Arabia.
⭐ 20 handcrafted features capturing signal strength dynamics, signal stability, channel characterization, and mobility metrics enable fine-grained classification.
⭐ Our XGBoost-based model achieves an average F1-score of 93.7%, outperforming the state-of-the-art (AutoSense) by more than 20% under a user-independent setting.
🌱 Future Directions:
Extending the framework to assess robustness across varying climatic conditions and to newer network generations such as 5G NR, as well as exploring uplink-based inference for real-time, network-native deployment.
📄 Please check out the paper for more details.
Harnessing Cellular Network Data for Accurate Transportation Mode Recognition | IEEE Journals & Magazine | IEEE Xplore
#TransportationModeRecognition #CellularData #MachineLearning #XGBoost #SmartMobility #5G
🔥 In this work, we show that transportation modes — bus, car, walk, and train — can be accurately classified using only radio access network (RAN) measurements from cellular infrastructure, without relying on GPS, IMU sensors, or any user-installed apps.
💡 Key Highlights:
⭐ A privacy-preserving, energy-efficient framework that uses only 4G cellular measurements from serving and neighboring cells, designed for operator-side deployment.
⭐ We introduce CellMob, a large-scale dataset of 160 trips and 55+ hours of travel time across two urban cities (Makkah, Jeddah) and a semi-rural campus (KAUST) in Saudi Arabia.
⭐ 20 handcrafted features capturing signal strength dynamics, signal stability, channel characterization, and mobility metrics enable fine-grained classification.
⭐ Our XGBoost-based model achieves an average F1-score of 93.7%, outperforming the state-of-the-art (AutoSense) by more than 20% under a user-independent setting.
🌱 Future Directions:
Extending the framework to assess robustness across varying climatic conditions and to newer network generations such as 5G NR, as well as exploring uplink-based inference for real-time, network-native deployment.
📄 Please check out the paper for more details.
Harnessing Cellular Network Data for Accurate Transportation Mode Recognition | IEEE Journals & Magazine | IEEE Xplore
#TransportationModeRecognition #CellularData #MachineLearning #XGBoost #SmartMobility #5G
