Uncertainty is considered as one of the most important limitations for applying the results of artificial intelligence techniques (AI) in water quality management to obtain appropriate control strategies. In this research, a proper methodology was proposed to determine the uncertainty of support vector machine (SVM) for the prediction of five-day biochemical oxygen demand (BOD5). In this regard, the SVM model was calibrated using different records for many times (here, 1000 times), to investigate model performance according to calibration pattern changes. Therefore, to implement the random selection of calibration patterns for several times, an alternative database was required. By this methodology, the parameters of SVM model will be obtained 1000 times, giving various predicted BOD5 values each time. To evaluate the SVM model's uncertainty, the percentage of observed data bracketed by 95 percent predicted uncertainties (95PPU) and the band width of 95 percent confidence intervals (d-factor) were selected. Findings indicated that the SVM model was more sensitive to capacity parameter (C) than to kernel parameter (Gamma) and error tolerance (Epsilon). Besides, results showed that the SVM model had acceptable uncertainty in BOD5 prediction. It is notified that the novelty of the presented methodology is beyond a mere application to water resources, and can also be used in other fields of sciences and engineering.
- Support vector machine
- Uncertainty analysis