Designing a Hybrid Machine Learning Model for Weather Forecasting in Batam City

Accurate weather forecasting in tropical regions such as Batam City is challenging due to high climate variability and frequent data gaps caused by unstable atmospheric conditions. This study aims to develop a reliable daily average temperature forecasting system using a hybrid approach that combines the Seasonal Autoregressive Integrated Moving Average (SARIMA) model and the Long Short-Term Memory (LSTM) neural network. The main novelty of this research lies in the residual hybridization method, where SARIMA is used to capture linear seasonal patterns and LSTM is applied to model the non-linear residual components, as well as the use of a multi-source data integration strategy to fill missing data. Historical temperature data from BMKG and other publicly available meteorological sources were merged to produce a continuous dataset covering the period from 2015 to 2021. The study evaluated several model architectures, including standalone statistical models, standalone machine learning models, and hybrid models, to identify the most effective approach. The experimental results show that the SARIMA–LSTM hybrid model outperformed the other models, achieving a high prediction accuracy with an R² value of 0.92 and a Root Mean Square Error (RMSE) of 1.73°C. These findings demonstrate that integrating linear and non-linear models can significantly improve temperature forecasting performance and provide a practical solution for weather monitoring in tropical environments