World Journal of Engineering Research and Technology

*Isaac Raphael Okafor, Garfield Jones, Guangming Chen

ABSTRACT

The importance of feature selection cannot be underrated due to its significance in dsetermining performance and reliability of predictive models. Machine learning techniques such as extratrees (ET) and random forest (RF) models were developed to leverage feature importance selection in predicting daily solar radiation (DSR) with precision. Daily solar radiation data for Southeastern (SE) Nigeria, such as temperature, relative humidity, precipitation, and wind speed were collected from the National Aeronautics and Space Administration (NASA) Prediction of Worldwide Energy Resource (POWER) database over a 10-year period ranging from 2012 to 2021 for the purpose of model development and testing. The findings reveal high impact of temperature and rank most important feature for the prediction of DSR by ET and RF models respectively in the five states.Among other features by the models, the temperature results are given as Abia- ET: Temp 0.275 and RF: Temp 0.267; Anambra- ET: Temp 0.278 and RF: Temp 0.293; Ebonyi- ET: Temp 0.285 and RF: Temp 0.266; Enugu- ET: Temp 0.274 and RF: Temp 0.267; Imo- ET: Temp 0.277 and RF: Temp 0.276. These results demonstrate great correlations in temperature values between the ET and RF models in the prediction of feature importance.

[Full Text Article] [Download Certificate]