Table 2. Parameters and statistics of the best selected architecture for each proposed framework, in order to predict the Bitcoin price, are described.
Statistics represent the average and standard deviation (in brackets), across the k-folds with k = 3, of the MAPE values. Note that (r) stands for SVR with a radial kernel function, (l) stands for SVR with a linear kernel function, and (p) stands for SVR with a polynomial kernel function. The bold entries highlight the framework with the lowest average of the MAPE’s values across the k-folds.
| LSTMNN | SVR (r) + LSTMNN | SVR (l) + LSTMNN | SVR (p) + LSTMNN | |
|---|---|---|---|---|
| # epochs | 83 | 330 | 319 | 136 |
| # neurons | 50 | 22 | 99 | 93 |
| # batchs | 40 | 96 | 87 | 98 |
| γ | 11.53 | |||
| d | 1 | |||
| avg (std) | 5.79 (1.00) | 6.76 (1.56) | 2.66 (0.08) | 2.7 (0.31) |