Metode Hibrida ARIMAX-SVR dengan Optimasi Hiperparameter Berbasis Particle Swarm Optimization untuk Peramalan Konsentrasi PM2.5
Dara Oktavia Risti, Prof. Dr. Drs. Gunardi, M.Si.
2026 | Skripsi | STATISTIKA
The high concentration of PM2.5 as one of the major air pollutants has become an environmental issue that receives increasing attention in many urban areas, including Yogyakarta, due to its adverse impacts on public health and environmental quality. PM2.5 concentration is influenced by various meteorological factors and anthropogenic activities, resulting in complex, dynamic, and nonlinear time series patterns. The Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX) model has limitations in modeling nonlinear patterns that remain in the residuals. On the other hand, Support Vector Regression (SVR) is capable of modeling nonlinear relationships, but its performance is highly dependent on the appropriate selection of hyperparameters.
In this study, hourly PM2.5 concentration forecasting in Yogyakarta was conducted using a hybrid ARIMAX–SVR method by incorporating exogenous variables, namely air temperature, relative humidity, and wind speed. Hyperparameter optimization for the SVR model was performed using the Particle Swarm Optimization (PSO) algorithm. The performance of the proposed method was then compared with the ARIMAX model, the SVR model, and the hybrid ARIMAX–SVR model without optimization using the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) evaluation metrics.
The results indicate that the PSO-optimized hybrid ARIMAX–SVR model achieved the lowest RMSE and MAE values among the competing models, thereby providing the best forecasting performance. Therefore, PSO-based hyperparameter optimization is shown to improve the forecasting accuracy of the hybrid ARIMAX–SVR model for PM2.5 concentration data in Yogyakarta.
Kata Kunci : PM2.5, ARIMAX, Support Vector Regression (SVR), Particle Swarm Optimization (PSO), peramalan.