Evaluation Of Time Series Forecasting Techniques For Air Quality Prediction: Case Study of NO₂ Levels

Air quality is an important factor for human life, as a decrease in air quality has a major impact on human health. Nitrogen dioxide (NO2) is one of the main pollutants that has a direct effect on human health. This study assesses the performance of three models for time series forecasting, namely AutoRegressive Integrated Moving Average (ARIMA), eXtreme Gradient Boosting (XGBoost) and Long Short-Term Memory (LSTM) in predicting NO2 concentration. The data used is air quality data from Beijing over the period 2013 to 2024. The research methodology includes data pre-processing steps to address missing values and reduce noise. Afterwards, ARIMA, XGBoost and LSTM models are applied, with the most maximized parameters selected through a cross-validation process. The XGBoost model based on the test results is the model with the best performance among other models, evidenced by the lowest prediction error values (RMSE, MAPE, MAE). The ARIMA and LSTM models, although providing good results, are still below the performance of XGBoost, XGBoost is able to capture fluctuations in NO2 concentration more accurately. This finding suggests that XGBoost is superior in modeling the complexity of air quality time series data. With the results obtained, it is expected to contribute to the development of better air quality prediction strategies and support more informed decision making for public health improvement.