<![CDATA[This study aims to analyze and compare the performance of two conventional time series forecasting models, Holt–Winters and Seasonal Autoregressive Integrated Moving Average (SARIMA), in predicting monthly climate variables in Semarang City, including air temperature, rainfall, and humidity. A head-to-head multivariable comparison was conducted within a single experimental framework in a tropical coastal climate context. Daily climate data from February 2017 to December 2023 were obtained from Kaggle and preprocessed through data completeness checks, date format conversion, and aggregation into monthly series. The dataset was divided into 80% training data and 20% testing data. In addition to static data splitting, a time series validation approach was applied to assess the stability of model performance across different training windows. Forecasting accuracy was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results indicate that Holt–Winters and SARIMA achieve comparable performance for temperature and humidity variables, with MAPE values of approximately 2.12% and 7.76%, respectively. In contrast, both models exhibit substantially higher errors when applied to rainfall data due to strong fluctuations and the presence of extreme values, indicating serious and shared limitations in accurately forecasting rainfall. This study concludes that conventional time series methods are effective for climate variables with relatively stable seasonal patterns but have inherent limitations in modeling highly volatile variables such as rainfall in tropical coastal regions.]]>
Copyrights © 2026
