Table 5 Comparison of the forecasting performance achieved with related works stratified by forecasting horizon
Forecasting Horizon | Reference and year | Method(s) used | MAPE(%) | RMSE |
|---|---|---|---|---|
3 up to 7 days ahead | Marcilio et al. [42] 2013 | GLM and GEE | 4,5–9,9 | not applied |
Xu [18] 2016 | ARIMA-LR (smoothing), ARIMA-LR and GLM | 6,8–9,6 | 70,5–104 | |
Calegari et al. [46] 2016 | SARIMA, SS and SMHW | 10,67–12,01 | not applied | |
Asheim et al. [48] 2019 | Poisson time-series regression model | 31–38 | not applied | |
Jilani et al. [15] 2019 | NN e FTS | 3.03–7,42 | 6,16–16,55 | |
Whitt et al. [3] 2019 | SARIMAX | 8,4–10,59 | not applied | |
Zhang et al. [35] 2019 | ARIMA, SVR and ARIMA-SVR | 7,02–7,36 | 19,20–20,34 | |
Choudhury and Urena [1] 2020 | ARIMA, HW and NN | not applied | 1,55–27,86 | |
Yousefi et al. [21] 2020 | LSTM | 5,59–6,31 | not applied | |
Erkamp et al. [6] 2021 | MLR | 8.68–12.20 | not applied | |
Rocha and Rodrigues [5] 2021 | RNN, XGBoost and RNN-XGBoost | not applied | 4,7–4,9 | |
Vollmer et al. [20] 2021 | GLMNET, LM and GBM | 6,7–8,6 | not applied | |
Sudarshan et al. [7] 2021 | RF, LSTM and CNN | 8,91–10,69 | not applied | |
Cheng et al. [61] 2021 | SARIMAX, HW and VAR | 5–15,3 | not applied | |
Murtas et al. [83] 2022 | ARIMA | 6,6–11,2 | not applied | |
Petsis et al. [40] 2022 | XGBoost | 6,5–6,91 | 22,96–23,9 | |
Tuominen et al. [54] 2022 | ARIMAX, RLS-FS and RLS-SA | 6,6–6,9 | not applied | |
Tello et al. [57] 2022 | ARIMA and SVR | 3,34–5,17 | 14,10–20,57 | |
Zhang et al. [41] 2022 | SVR, RF and KNN | 8,81–9,63 | 26,84–30,23 | |
Current study 7-day test set | ED ARMA | GLMNET and SVM-RBF | 5,48–5,52 | 11,44–11,69 |
ED JOON | SVM-RBF and LightGBM | 4,61–4.73 | 15,95–16,22 | |
ED RG | RF and LightGBM | 6,55–6,81 | 11,93–12,87 | |
ED RPH | XGBoost and SVM-RBF | 5,90–6,21 | 14,89–15,45 | |
ED SCG | XGBoost and NNAR | 5,08–5,21 | 11,73–11,93 | |
8 up to 45 days ahead | Marcilio et al. [42] 2013 | GLM and GEE | 8,7–12,8 | not applied |
Bergs et al. [45] 2014 | ETS | 2,63–4,76 | not applied | |
Calegari et al. [46] 2016 | SARIMA, SS and SMHW | 11,35–12,29 | not applied | |
Juang [19] 2017 | ARIMA | 8,91 | not applied | |
Carvalho-Silva et al. [23] 2018 | ARIMA | 5,22–9,29 | not applied | |
Jilani et al. [15] 2019 | NN e FTS | 2,01–2,81 | 57,30–167,89 | |
Khaldi et al. [27] 2019 | EEMD-ANN, DWT-ANN and ANN | not applied | 52,86–149,23 | |
Vollmer et al. [20] 2021 | GLMNET, LM and GBM | 6,8–8,9 | not applied | |
Pekel et al. [39] 2021 | PSO-ANN, Bayesian ANN and GA-ANN | 6–8,8 | 53.29–83.85 | |
Tuominen et al. [54] 2022 | ARIMAX, RLS-FS and RLS-SA | 7,4–7,8 | not applied | |
Susnjak et al. [78] 2023 | Voting regressor | 8,9–12,8 | 10,60–15,9 | |
Gafni-Pappas et al. [50] 2023 | RF and XGBoost | not applied | 18,94–18,96 | |
Current study 45-day test set | ED ARMA | XGBoost and GLMNET | 5,90–5,90 | 12,62–12,64 |
ED JOON | SVM-RBF and RF | 4.98–5.01 | 17,25–17,43 | |
ED RG | XGBoost and RF | 6,23–6,28 | 11,42–11,47 | |
ED RPH | SVM-RBF and NNAR | 5,71–5,85 | 14,25–14,37 | |
ED SCG | XGBoost and RF | 5,64–5,69 | 12,61–12,56 |