<![CDATA[Air temperature is a critical variable in weather conditions that affects various aspects of human life, including health, agriculture, and the economy. In Indonesia, particularly in Mataram City, which is situated in a tropical region, significant temperature changes can impact sectors such as tourism, agriculture, and daily activities. Accurate temperature forecasting can aid the public, industries, and the government in making more informed decisions, both for short-term and long-term planning. However, weather in tropical regions like Mataram tends to be difficult to predict accurately due to its dynamic nature and the influence of multiple atmospheric factors. Conventional weather prediction methods often fail to capture the complex patterns in historical temperature data, necessitating more advanced methods to improve forecast accuracy. Recurrent Neural Networks (RNNs), particularly the Long Short-Term Memory (LSTM) variant, have proven to be highly effective tools for modeling complex time series data. This algorithm can retain long-term information and recognize patterns in data that change over time, making it well-suited for temperature prediction challenges. In this study, the RNN-LSTM algorithm is applied to forecast temperatures in Mataram City, aiming to improve forecast accuracy and produce results useful for various purposes. The temperature prediction model using the LSTM algorithm involves several steps: data collection, data normalization, splitting data into test and training sets, building the LSTM model by determining the number of epochs, layers, and batch size, and finally, evaluating the model with RMSE. Two parameters, epoch and batch size, influence the LSTM model’s forecasting results in this study. Epochs used in this study are 5, 10, 20, 30, 40, 50, and 100, with a fixed batch size of 32. The LSTM algorithm employs the RMSProp optimizer. The temperature prediction model using the LSTM method achieved the best average accuracy with a batch size of 32 and 50 epochs, yielding an RMSE value of 0.13 and a prediction accuracy of 99.96% in forecasting Mataram City’s temperature for the year 2023.]]>
