This study presents an air temperature forecasting model employing the Long Short-Term Memory (LSTM) algorithm to address the challenges posed by climate variability and extreme weather conditions. Historical daily temperature data from NASA POWER—collected between January 1, 2014, and December 31, 2024, in Serang City (totaling 4,018 records)—were used. The data were normalized using a min–max scaling technique and divided into training (70%) and testing (30%) sets. Multiple experimental scenarios were run by varying the number of training epochs and the hidden layer unit counts. The optimal configuration was achieved in Scenario 7, which incorporated two hidden layers, each with 50 units, and employed 30 epochs; this setup yielded a prediction accuracy of 98.4% with a Root Mean Squared Error (RMSE) of 27.11. The results indicate that the LSTM model effectively captures the seasonal variations and long-term trends in air temperature, making it a reliable tool for forecasting and supporting decision-making in climate adaptation strategies. Keywords: Air Temperature Forecasting, Long Short-Term Memory, Deep Learning, Climate Change, Data Normalization.