Skip to main content

Table 2 Comparison of the performance of the different ML methods

From: A comprehensive study on battery electric modeling approaches based on machine learning

ML method RMSE MAE Tr. time Test. time Memory Npar/
  in [mV] in [mV] in [s] in [s] in [MB] in [-]
MLR 48.0 37.7 0.02 <0.01 15.4 6
SVR 18.3 10.1 1676 51.0 2.82 28048
K-NN 37.9 22.3 0.50 7.10 17.2 2.2 Mio.
Decision Tree 14.6 8.34 1.89 0.01 2.56 29998
Random Forest 11.7 6.48 24.4 0.22 11.5 0.6 Mio.
AdaBoost 10.8 5.98 61.7 0.59 29.8 1.2 Mio.
GBR 10.4 5.29 6.92 3.29 31.3 43681
FFNN 5.57 3.51 69.7 0.15 3.51 5041
MergeFFNN 7.63 4.75 78.2 0.22 3.54 8011
MergeLSTM 8.45 5.66 1307 2.89 4.15 1201
LSTM 8.24 5.46 1793 4.87 4.15 1821
MergeCNN 7.39 4.73 194 0.68 3.76 5651
CNN 7.99 5.21 273 0.80 3.75 7511
  1. The number of parameters Npar for model parametrization is derived from the optimized model architecture