Skip to main content

Table 2 Coverage of traffic forecasting methodologies by FSE methods

From: Feature selection and extraction in spatiotemporal traffic forecasting: a systematic literature review

Model Short description Total number of applications Number of applications of FSE methods’ class
Exogenous feature filtering Endogenous feature filtering Wrapper feature selection Embedded feature selection Dimension reduction
Artificial neural networks (ANN)
 FFNN Feedforward ANN 41 34 4 9 6
 TDNN Time-delayed ANN 9 13 2 2
 RNN Recurrent ANN 9 13 2 2
 LSTM Long short-term memory ANN 6 3 2 1
 SSNN State-space ANN, incl. Time-delayed state-space ANN (STDNN) 4 4 2
 CNN Convolutional ANN 4 4
 DBN Deep belief network, incl. Restricted Boltzmann machine (RBM), stacked autoencoders (SAE), generative adversarial networks (GAN) 4 1 4
 NARX Nonlinear autoregressive exogenous ANN 3 3
 Other NN architectures Incl. counter-propagation ANN (CPNN), fuzzy ANN, Graph ANN, general regression ANN, group method of data handling (GMDH) 8 4 1 2 1
Statistical models
 BN Bayesian networks, incl. Conditional random fields 30 15 8 2 1 6
 DL/ARDL Distributed lags /autoregressive distributed lag models, incl. Smoothing models, chaos models 26 19 8 2 2 1
 SVR Support vector regression, incl. Extreme learning machine 23 8 2 4 2 11
 KNN k-nearest neighbour regression 22 15 2 4 0 3
 VAR Vector autoregressive models 21 22 5 2
 STARIMA Space-time autoregressive integrated moving average, incl. Generalised STARIMA 21 20 13 1
 Kernel Kernel regressions, incl. Gaussian process regression (GPR) 10 7 4
 State-space State-space models 10 8 2
 Tensor models Tensor completion models, incl. Probabilistic principal component analysis (PPCA) 6 1 6
 Decision tree models Incl. random forest and regression tree 5 4 2 1 1
 SCTM Stochastic cell transmission model 4 3 2
 MARS Multivariate adaptive regression splines, incl. Generalised additive model (GAM) 3 1 2
 Spatial panel Spatial panel models 2 1 2