Using Thai channel measurements, classifiers predicted whether NB-IoT would achieve 95% RSRP coverage across 411 cells at four sites. Gradient Boosting reached 0.971 accuracy, 0.969 F1 and 1.7 ms inference, far faster than Monte Carlo. However, training labels came from Monte Carlo under the same assumed channel model, so this emulates simulation rather than proving field-deployment accuracy.
Key findings
- Gradient Boosting achieved 0.971 accuracy, 0.969 F1 and 1.7 ms latency, about four orders faster than direct simulation. Under the model assumptions, rural Suphan Buri was the only RECOMMENDED case, with 88.5% feasible cells, while hybrid PLC backhaul was suggested for dense urban areas.
Why this matters globally
A rapid screening tool could help utilities prioritise field surveys and compare NB-IoT with PLC before investment, avoiding repeated heavy simulations for every cell.
Thai researcher contribution
Kittiwat Srivilas and Chaiyod Pirak of TGGS-KMUTNB linked Thai channel measurements to Advanced Metering Infrastructure planning in the Provincial Electricity Authority context.
Limitations to consider
Accuracy is against simulated labels, not deployed-meter ground truth. Meter density is synthetic, only 411 cells at four sites and one threshold were used, and spatial leakage is possible unless sites were held out. The provincial recommendation is a scenario, not an investment decision.
Verify the original sources
EnergiesRead the original article↗DOI: 10.3390/en19133195