Table 3.
Classification accuracy in held out simulated systems. While the datasets in Tables 1, 2 were used to optimize the Chaos Decision Tree Algorithm, these datasets were not. Performance was near-perfect.
| Measurement noise level (% of std. dev.) | |||||
|---|---|---|---|---|---|
| System | 0% | 10% | 20% | 30% | 40% |
| Rössler system59 (chaotic) | 70/100 | 90/100 | 96/100 | 100/100 | 100/100 |
| Ikeda map60 (chaotic) | 100/100 | 100/100 | 100/100 | 100/100 | 14/100 |
| Hénon map61 (periodic) | 100/100 | 100/100 | 100/100 | 100/100 | 100/100 |
| Cubic map61 (period-doubled) | 100/100 | 100/100 | 100/100 | 100/100 | 100/100 |
| Poincaré oscillator41 (periodic) | 100/100 | 100/100 | 100/100 | 100/100 | 57/100 |
| Poincaré oscillator41 (quasi-periodic) | 100/100 | 100/100 | 100/100 | 100/100 | 100/100 |
| Poincaré oscillator41 (chaotic) | 100/100 | 100/100 | 100/100 | 100/100 | 100/100 |
| Multivariate AR model (linear stochastic) | 100/100 | 100/100 | 100/100 | 100/100 | 100/100 |