Comparing SVR and Random Forest Forecasting based on Autoregressive Time Series with Application
Abstract
The accuracy of forecasting the time series of relative humidity in its maximum and minimum cases is important for controlling environmental impacts, damages and risks. In this study, the support vector regression (SVR) method and the random forest (RF) method will be used, depending on the principle of auto regressive (AR) and the autocorrelation (AC), which is the main characteristic of time series in general. The Lags of original time series will be depended as the explanatory (input) variables while the original series will be as target variable. This structure is fitted with the AC principle because the current observation will be depending on time lags in each time step of time series variable. Comparisons of the forecasting results will be performed by using SVR , RF methods and compared to the classical method of analysing time series which is the integrated autoregressive and moving average (ARIMA) model. The SVR and RF methods were employed due to their importance in improving the forecast results, as they are the ideal solution to the problem of non-linearity of the data, as well as the problem of heterogeneity in the climate data, especially as a result of the fact that they contain many seasonal and periodic compounds, which may lead to inaccurate forecast. The forecast of the time series of relative humidity in its minimum and maximum cases was studied in this study for one of the agricultural meteorological stations in the city of Mosul-Iraq. The results of this study reflected the superiority of both SVR method and RF method compared to the classical method represented by the ARIMA model. The results also included the superiority of the RF method in forecasting the training period compared to the SVR method, which was more balanced despite that, as it superiority the results of ARIMA in forecasting the training period and the testing period, while it was its forecast performance is slightly better than the forecast results of the RF method in the test period.
References
- Ahmed, H.U., et al. Support vector regression (SVR) and grey wolf optimization (GWO) to predict the compressive strength of GGBFS-based geopolymer concrete. Neural Computing and Applications. 2023; 35(3): 2909-2926.
- Al-Musaylh, M.S., et al. Two-phase particle swarm optimized-support vector regression hybrid model integrated with improved empirical mode decomposition with adaptive noise for multiple-horizon electricity demand forecasting. Applied energy. 2018; 217: 422-439.
- Breiman, L. Random forests. Machine learning. 2001; 45: 5-32.
- Chaudhuri, S. and D. Dutta. MannKendall trend of pollutants, temperature and humidity over an urban station of India with forecast verification using different ARIMA models. Environmental monitoring and assessment. 2014; 186: 4719-4742.
- Cortes, C. and V. Vapnik. Support-vector networks. Machine learning. 1995; 20: 273-297.
- Das, R.C. Forecasting incidences of COVID-19 using Box-Jenkins method for the period July 12-Septembert 11, 2020: A study on highly affected countries. Chaos, Solitons & Fractals. 2020; 140: 110248.
- Eymen, A. and . Kyl. Seasonal trend analysis and ARIMA modeling of relative humidity and wind speed time series around Yamula Dam. Meteorology and Atmospheric Physics. 2019; 131: 601-612.
- Hyndman, R.J. and A.B. Koehler. Another look at measures of forecast accuracy. International journal of forecasting. 2006; 22(4): 679-688.
- Kumar, M. and M. Thenmozhi. Forecasting stock index returns using ARIMA-SVM, ARIMA-ANN, and ARIMA-random forest hybrid models. International Journal of Banking, Accounting and Finance. 2014; 5(3): 284-308.
- Li, Z.-C., et al. Prediction of protein structure class by coupling improved genetic algorithm and support vector machine. Amino acids. 2008; 35: 581-590.
- Noureen, S., et al. A comparative forecasting analysis of arima model vs random forest algorithm for a case study of small-scale industrial load. International Research Journal of Engineering and Technology. 2019; 6(09): 1812-1821.
- Shukur, O.B. and M.H. Lee. Daily wind speed forecasting through hybrid KF-ANN model based on ARIMA. Renewable Energy. 2015; 76: 637-647.
- Shumway, J., et al. Machine learning to improve the prioritization and effectiveness of pre-treatment physics chart checks. International Journal of Radiation Oncology, Biology, Physics. 2020; 108(3): S54-S55.
- Xu, D., et al. Application of a hybrid ARIMASVR model based on the SPI for the forecast of droughta case study in Henan Province, China. Journal of applied meteorology and climatology. 2020; 59(7): 1239-1259.
- Zhang, F. and L.J. O'Donnell, Support vector regression, in Machine learning. 2020, Elsevier. p. 123-140.
