Official Journal of the Asia Oceania Geosciences Society (AOGS)
Table 5 Univariate time series forecasting methods of this study (part 2): machine learning methods
From: One-step ahead forecasting of geophysical processes within a purely statistical framework
s/n | Abbreviated name | Category | Model structure information | R algorithm(s) | Implementation notes | |
|---|---|---|---|---|---|---|
Hyperparameter optimized using grid search (grid values) | Lagged variable selection procedure (see Table 6) | |||||
12 | NN_1 | Neural networks | Single hidden layer multilayer perceptron | CasesSeries {rminer}, fit {rminer}, lforecast {rminer}, nnet {nnet} | Number of hidden nodes (0, 1, …, 15) | 1 |
13 | NN_2 | 2 | ||||
14 | NN_3 | nnetar {forecast} | 3 | |||
15 | RF_1 | Random forests | Breiman’s random forests algorithm with 500 grown trees | CasesSeries {rminer}, fit {rminer}, lforecast {rminer}, randomForest {randomForest} | Number of variables randomly sampled as candidates at each split (1, …, 5) | 1 |
16 | RF_2 | 2 | ||||
17 | RF_3 | 3 | ||||
18 | SVM_1 | Support vector machines | Radial basis kernel “Gaussian” function, C = 1, epsilon = 0.1 | CasesSeries {rminer}, fit {rminer}, lforecast {rminer}, ksvm {kernlab} | Sigma inverse kernel width (2n, n = − 8, − 7, …, 6) | 1 |
19 | SVM_2 | 2 | ||||
20 | SVM_3 | 3 | ||||