A set of metrics for the effective evaluation of point forecasting methods used for hydrological tasks

G. Papacharalampous, H. Tyralis, and D. Koutsoyiannis, A set of metrics for the effective evaluation of point forecasting methods used for hydrological tasks, Asia Oceania Geosciences Society (AOGS) 14th Annual Meeting, Singapore, HS01-A001, doi:10.13140/RG.2.2.19852.00641, Asia Oceania Geosciences Society, 2017.

[doc_id=1718]

[English]

The selection of metrics for the evaluation of point forecasting methods can be challenging even for very experienced hydrologists. We conduct a large-scale computational experiment based on simulations to compare the information that 18 metrics proposed in the literature give about the forecasting performance. Our purpose is to provide generalized results; thus we use 2 000 simulated Autoregressive Fractionally Integrated Moving Average time series. We apply several forecasting methods and we compute the values of the metrics for each forecasting experiment. Subsequently, we measure the correlation between the values of each pair of metrics, separately for each forecasting method. Furthermore, we explore graphically the detected relationships. Finally, we propose a set of metrics that we consider to be suitable for the effective evaluation of point forecasting methods.

PDF Full text: