Prediction of Daily Dewpoint Temperature Using a Model Combining the Support Vector Machine with Firefly Algorithm

In this research work, a hybrid approach of integrating a support vector machine (SVM) with firefly algorithm (FFA) is proposed to predict daily dewpoint temperature ($T_{\text{dew}}$). The main aim of employing FFA is to identify the optimal SVM parameters and provide the possibility of enhancing the SVM’s capability. The weather data sets including 10 years of measured-daily average air temperature ($T_{\text{avg}}$), relative humidity ($R_h$), atmospheric pressure ($P$), and $T_{\text{dew}}$ for an Iranian city have been utilized. Seven different sets of parameters with one, two, and three of the considered parameters serve as inputs to establish seven models. The capability of the SVM-FFA method is compared against SVM, artificial neural network (ANN), and genetic programming (GP) to demonstrate its efficiency and viability. It is found that further precision is achieved for Model 7 established based on all approaches utilizing three inputs of $T_{\text{avg}}$, $R_h$, and $P$. The obtained results clearly indicate that the SVM-FFA method, by providing very favorable predictions, outperforms other examined techniques. In fact, hybridizing the SVM with FFA can be particularly promising as it favorably enhances the SVM’s accuracy. For the SVM-FFA Model 7, as the best model, the mean absolute bias error, root mean square error, and correlation coefficient obtained are equal to 0.6863°C, 0.8959°C, and 0.9849, respectively. While for the SVM Model 7, ranked in the next place, the attained values are 0.8810°C, 1.1487°C, and 0.9760, respectively. In summary, the SVM-FFA is indeed effective to predict daily $T_{\text{dew}}$ with greater precision and reliability.