Sistem Prediksi Kualitas Udara Menggunakan Algoritma Long Short-Term Memory (LSTM)

Abstract

Conventional and static air quality monitoring in Yogyakarta City which only presents historical (past) data reports  hinders proactive mitigation efforts against air pollution. This research aims to develop an air quality prediction system using the Long Short-Term Memory (LSTM) algorithm, a deep learning method superior for time-series data analysis. The system utilizes historical data from the Yogyakarta City Environmental Agency (DLH) from 2022 to 2024, covering pollutant parameters such as PM10, PM2.5, SO₂, CO, O₃, and NO₂. The primary prediction focus is the AQI (Air Quality Index) value, calculated based on the concentration of these pollutant parameters.  The research method includes data preprocessing, such as handling missing data with interpolation, designing a two-layer LSTM model architecture, model training, and performance evaluation using Mean Absolute Error (MAE) and Mean Absolute Deviation (MAD) metrics. The results show that the developed LSTM model successfully provides predictions with good performance, where the combined average MAE value (4.85) is significantly lower than the average MAD of the actual data (10.19), indicating that the model's prediction error is smaller than the natural variability of the data. The output of this research is a prototype application with a graphical user interface (GUI) capable of displaying air quality predictions for the next day, identifying critical pollutant components, and presenting air quality condition classifications informatively.