International Journal of Environment and Climate Change

This study trained six machine learning models to predict meteorological variables at a tropical location. The models used are: Multiple linear regression, Decision tree, Random forest, Support vector machine, Extreme gradient boosting and Multilayer perceptron. This was with the aim of determining the best machine learning model for weather forecasting in a tropical location. The meteorological variables that were predicted are: Temperature, Solar radiation, Relative humidity and Wind speed.  To identify the efficiency and to quantify the predictive capacity of each models, evaluation metrics such as coefficient of determination (R²), mean absolute error (MAE), mean absolute percentage error (MAPE) and mean square error (RMSE) were employed. The best performed model for temperature is the Random Forest which has R² of 0.93, MAE of 0.78 ⁰C, MAPE of 2.84 % and RMSE of 1.13 ⁰C. Also, the best performed model for solar radiation is the Random Forest having an R² value of 0.72, MAE of 85.34 W/m² and RMSE of 19008.45 W/m². For relative humidity, Random Forest also has the best performance. From the evaluation metrics, it has R² of 0.92, MAE of 3.41 %, MAPE of 0.75 % and RMSE of 24.71 %. The best performed technique for predicting the wind speed was also the Random Forest having an R² value of 0.79, MAE of 0.16 m/s and RMSE of 0.044 m/s. The study concluded that the best machine learning model for predicting meteorological variables in a tropical location is the Random Forest.