Machine Learning-Based Temperature and Humidity Prediction for a Coastal Area: A Case Study

Human lives and livelihoods are being adversely affected by hazards caused by climate change. Timely broadcasts of accurate meteorological information can reduce these hazards. Among all the climatic attributes, the prediction of temperature and relative humidity plays a vital role in improving the accuracy of weather forecasting. This paper aims to forecast temperature and relative humidity using four different machine learning algorithmsâ Lasso, Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). Daily data for 21 years was obtained from Teknaf weather station, Bangladesh Meteorological Department (BMD). The findings demonstrated that ensemble-based models outperform the other models in accurately forecasting daily temperature and humidity within a reasonable range. The RF model was the most accurate for temperature prediction, achieving a mean squared error (MSE) of 0.645°C, a mean absolute error (MAE) of 0.611°C, a mean absolute percentage error (MAPE) of 2.357%, and a coefficient of determination R2 of 0.933. Moreover, for humidity prediction, the XGB model outperformed the others, with an MSE of 18.619%, MAE of 3.275%, MAPE of 4.305%, and R2 of 0.720. These results suggest that ensemble-based models, particularly RF and XGB, offer higher accuracy in forecasting temperature and relative humidity compared to other methods.