Analisis Model Long Short-Term Memory (LSTM) untuk Prediksi Kadar Particulate Matter 2.5

Abstract

Particulate Matter 2.5 (PM2.5) is a hazardous air pollutant that can adversely affect human health and the environment. Accurate prediction of PM2.5 concentrations is essential to support air quality monitoring and pollution mitigation policy-making. This study aims to analyze the performance of the Long Short-Term Memory (LSTM) model in predicting PM2.5 concentrations based on time series data. The LSTM model, as a variant of Recurrent Neural Networks (RNN), is selected due to its ability to capture long-term dependencies in sequential data and to overcome the vanishing gradient problem commonly encountered in conventional RNN models. This study utilizes historical PM2.5 concentration data along with supporting variables such as temperature, humidity, wind speed, and other pollutant parameters. The research process includes data preprocessing, determination of the LSTM architecture (number of layers, hidden units, timestep), model training, and evaluation using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The results indicate that the LSTM model is capable of providing accurate predictions for short- to medium-term forecasting horizons, with performance superior to baseline models such as ARIMA and simple RNN. This study is expected to serve as a reference for the development of more reliable and practical deep learning-based air quality prediction systems in urban areas.