Table 5 Comparison of forecasting methods
From: Time series big data: a survey on data stream frameworks, analysis and algorithms
TM | Method | Year | PM | MV | Metrics | Notes |
|---|---|---|---|---|---|---|
ML | MELM, based on SARIMA, and on a top-K regression tree method | 2017 | [50] | Yes | ARMSE | Â |
| Â | FCCF, based on Bayes Networks and Gaussian Markov random fields | 2016 | [51] | Yes | RMSE | Â |
| Â | Based on LSSVM, and mixture kernels | 2022 | [48] | Yes | MSPE, MAPE, RMSE | Â |
| Â | Weighted combination of LightGBM models | 2021 | [40] | Yes | WRMSSE | The best method of the M5 competition. |
DL | DMVST-Net, based on LSTMs and CNNs | 2018 | [52] | Yes | MAPE, RMSE | The authors included a view for semantics. |
| Â | Hybrid approach of exponential smoothing with RNNs. | 2018 | [39] | Yes | sMAPE | The best method of the M4 competition. |
| Â | MTL-TCNN, based on Temporal Convolution, Convolution, and DTW | 2020 | [53] | Yes | MAE, MAPE, RMSE | Â |
| Â | Based on GCN and GRU | 2021 | [47] | Yes | MAE, MAPE, MSE | Drawbacks: The complexity of the method. |
| Â | Deep Neural Network | 2022 | [49] | Yes | \(\hbox {R}^{2}\), MAE, RMSE, MSE | Practical comparison of methods |
| Â | Based on CNNs, graphs, and LSTMs | 2022 | [46] | Yes | RMSE, MAPE | Â |