Table 2 Performance comparison of ML, DL, and time series models (lag = 5, output = 1).
From: Electric-load forecasting using interval models based on granularity and justifiable principles
Model | Test RMSE | Test MAE | Test \(R^2\) | Test MAPE | Train RMSE | Train MAE | Train \(R^2\) | Train MAPE |
|---|---|---|---|---|---|---|---|---|
Linear regression | 136.44 | 104.69 | 0.9872 | 2.34 | 136.74 | 104.16 | 0.9844 | 2.74 |
Decision tree | 174.27 | 131.85 | 0.9792 | 3.07 | 1.30 | 0.009 | 1.0000 | 0.00 |
SVM | 199.94 | 150.03 | 0.9726 | 3.32 | 165.70 | 131.81 | 0.9771 | 3.42 |
KNN | 135.13 | 101.27 | 0.9875 | 2.31 | 103.04 | 77.22 | 0.9912 | 2.03 |
Boosted ensemble | 133.52 | 99.68 | 0.9878 | 2.29 | 53.38 | 37.62 | 0.9976 | 0.98 |
Neural net | 146.54 | 111.19 | 0.9853 | 2.72 | 142.32 | 108.84 | 0.9831 | 2.48 |
LSTM | 129.78 | 100.44 | 0.9885 | 2.33 | 128.05 | 97.77 | 0.9863 | 2.58 |
Transformer | 170.78 | 137.93 | 0.9800 | 3.04 | 152.29 | 122.06 | 0.9807 | 3.08 |
ARIMA | 424.56 | 325.77 | 0.8765 | 7.11 | 404.60 | 311.64 | 0.8637 | 7.71 |
SARIMA | 399.40 | 303.75 | 0.8907 | 6.65 | 381.35 | 290.00 | 0.8789 | 7.21 |
