An XGBoost Machine Learning Technique for RF Energy Harvesting Prediction in IP-enabled IoT Devices

The number of interconnected devices in use world-wide continue to grow, placing a high demand on the limited energy available to power such devices. RF energy harvesting has been identified as a technique to mitigate the problem of energy availability, especially for wirelessly connected devices. it entails the conversion of RF energy from the ambient environment to DC energy in order to power host devices or charge their batteries. For IP-enabled IoT devices that have different subsections demanding electrical power, having a foreknowledge of the residual and harvestable energy is beneficial for optimum energy resource management. Machine Learning, which has become almost ubiquitous in its applications provides a capability for wireless IP-enabled IoT devices to predict harvestable energy from its environment. In this work, an Extreme Gradient Boost (XGBoost) machine learning model is investigated and compared with another machine learning model, Support Vector Regressor (SVR), using Normalized Root Mean Squared Error (nRMSE) and Mean Absolute Error (MAE) as performance metrics. Results obtained showed that the XGBoost model performed better than the SVR model across all datasets used in the investigation. More concisely, the XGBoost model showed an average nRMSE of 0.0727 across all datasets which is 7.43% lower than the SVR with an average nRMSE of 0.0785. Furthermore, both models benefited from increasing data size, with a decrease of 18.64% and 15.79% across the 1-day to 30-day datasets for the XGBoost and SVR, respectively.