Table 2.
Classification accuracy (mean ± standard deviation) [%] with five-fold cross-validation for the task of classifying material, gas, and ppm level using different set-ups. The number after DBN defines the window width (number of visible units) and the numbers after auto-encoder define the model order in the first and second layer.
| Material | Gas | ppm | Required input data | |
|---|---|---|---|---|
| SVM, 7 features | 89.0 ± 6.3 | 60.1 ± 11.1 | 42.9 ± 7.6 | 15 s to 14 min |
| SVM, 25 s raw data | 55.5 ± 3.7 | 53.8 ± 7.3 | 38.8 ± 4.1 | 25 s |
| DBN, 25 s | 86.8 ± 4.6 | 83.7 ± 4.1 | 49.5 ± 5.6 | 25 s |
| DBN, 10 s | 83.7 ± 1.9 | 61.4 ± 4.7 | 41.7 ± 6.8 | 10 s |
| DBN, 5 s | 72.4 ± 3.5 | 60.0 ± 4.5 | 33.0 ± 3.3 | 5 s |
| Auto-encoder, 5-5 | 93.2 ± 3.6 | 84.3 ± 3.8 | 61.2 ± 6.8 | 25 s |
| Auto-encoder, 5-2 | 95.7 ± 1.3 | 80.8 ± 6.6 | 51.1 ± 1.6 | 10 s |
| Auto-encoder, 3-2 | 91.8 ± 3.2 | 71.1 ± 5.8 | 45.3 ± 4.5 | 6 s |
| Auto-encoder, 2-2 | 61.7 ± 1.5 | 45.4 ± 4.7 | 36.5 ± 3.5 | 4 s |