Figure 1:
Figure 2:
Figure 3:
Figure 4:
Figure 5:
The evaluation performance of MLP
| Activation function | Hidden layer | Hidden node | Accuracy | Precision | Recall | F1-score |
|---|---|---|---|---|---|---|
| ReLU | 2 | 50 | 0.8472 ± 0.0461 | 0.8714 ± 0.0545 | 0.8472 ± 0.0461 | 0.8394 ± 0.0489 |
| 100 | 0.8750 ± 0.0461 | 0.8921 ± 0.0464 | 0.8750 ± 0.0461 | 0.8683 ± 0.0511 | ||
| 3 | 50 | 0.8611 ± 0.0278 | 0.8831 ± 0.0362 | 0.8611 ± 0.0278 | 0.8537 ± 0.0323 | |
| 100 | 0.8750 ± 0.0241 | 0.8921 ± 0.0361 | 0.8750 ± 0.0241 | 0.8703 ± 0.0246 | ||
| 4 | 50 | 0.8472 ± 0.0461 | 0.8564 ± 0.0581 | 0.8472 ± 0.0461 | 0.8399 ± 0.0494 | |
| 100 | 0.8472 ± 0.0241 | 0.8712 ± 0.0307 | 0.8472 ± 0.0241 | 0.8362 ± 0.0290 | ||
| Sigmoid | 2 | 50 | 0.7361 ± 0.0241 | 0.6907 ± 0.0486 | 0.7361 ± 0.0241 | 0.6816 ± 0.0385 |
| 100 | 0.7500 ± 0.0278 | 0.7738 ± 0.0636 | 0.7500 ± 0.0278 | 0.7149 ± 0.0247 | ||
| 3 | 50 | 0.7083 ± 0.0241 | 0.7401 ± 0.0736 | 0.7083 ± 0.0241 | 0.6486 ± 0.0305 | |
| 100 | 0.7222 ± 0.0000 | 0.7817 ± 0.0413 | 0.7222 ± 0.0000 | 0.6583 ± 0.0185 | ||
| 4 | 50 | 0.6806 ± 0.0461 | 0.5978 ± 0.1663 | 0.6806 ± 0.0461 | 0.5978 ± 0.0888 | |
| 100 | 0.7083 ± 0.0241 | 0.7153 ± 0.1564 | 0.7083 ± 0.0241 | 0.6191 ± 0.0495 | ||
Variation of the number of hidden layers, the number of hidden nodes, and the activation function of MLP architecture
| Activation Function | Hidden Layer | Hidden Node |
|---|---|---|
| ReLU | 2 | 50, 50 |
| 100, 100 | ||
| 3 | 50, 50 | |
| 100, 100 | ||
| 4 | 50, 50 | |
| 100, 100 | ||
| Sigmoid | 2 | 50, 50 |
| 100, 100 | ||
| 3 | 50, 50 | |
| 100, 100 | ||
| 4 | 50, 50 | |
| 100, 100 | ||
Sensor array used in the portable electronic nose and its specifications_
| Name | Specification |
|---|---|
| TGS 2602 | High sensitivity to gaseous air contaminants and VOCs such as ammonia, H2S, ethanol, and toluene. |
| TGS 2620 | High sensitivity to alcohol and organic solvent vapors and other volatile vapors, such as hydrogen, carbon monoxide, methane, and iso-butane. |
| MQ-7 | High sensitivity to carbon monoxide. |
| MQ-9 | High sensitivity to carbon monoxide, methane, and LPG. |
| TGS 2600 | High sensitivity to low concentrations of gaseous air contaminants such as hydrogen, carbon monoxide, methane, iso-butane, and ethanol. |
| TGS 2611 | High sensitivity and selectivity to methane gas, response to ethanol, hydrogen, and iso-butane. |
Hyperparameter settings for training using machine learning classifiers
| Classifier | Hyperparameter | Value |
|---|---|---|
| NB (type = multinomial NB) | alpha | [0, 0.1, 0.01, 0.001, 1.0] |
| fit_prior | [True, False] | |
| KNN | n_neighbors | 1–10 |
| P | [1, 2] | |
| weights | [‘uniform’, ‘distance’] | |
| SVM (type = SVC) | C | [1, 10, 100, 1000] |
| gamma | [0.1, 0.01, 0.001, 0.0001] | |
| kernel | [‘linear’, ‘poly’, ‘rbf’, ‘sigmoid’] | |
| DT | criterion | [‘gini’, ‘entropy’] |
| max_depth | [5, 10, 15, 20, 25, 30] | |
| max_features | [‘sqrt’, ‘log2’, None] | |
| min_samples_leaf | [2, 3, 4, 5] | |
| min_samples_split | [2, 3, 4, 5] | |
| splitter | [‘best’, ‘random’] | |
| RF | n_estimators | [10, 25, 50, 100] |
| criterion | [‘gini’, ‘entropy’, ‘log_loss’] | |
| max_depth | [5, 10, 15, 20, 25, 30] | |
| max_features | [‘sqrt’, ‘log2’, None] | |
| min_samples_leaf | [2, 3, 4, 5] | |
| min_samples_split | [2, 3, 4, 5] | |
The evaluation performance of machine learning classifiers
| Classifier | Accuracy | Precision | Recall | F1-score |
|---|---|---|---|---|
| NB | 0.5972 ± 0.0241 | 0.6513 ± 0.0877 | 0.5972 ± 0.0241 | 0.5432 ± 0.0181 |
| KNN | 0.7778 ± 0.0000 | 0.7897 ± 0.0253 | 0.7778 ± 0.0000 | 0.7575 ± 0.0112 |
| SVM | 0.7778 ± 0.0393 | 0.7792 ± 0.0473 | 0.7778 ± 0.0393 | 0.7702 ± 0.0433 |
| DT | 0.5972 ± 0.0992 | 0.6196 ± 0.0894 | 0.5972 ± 0.0992 | 0.6002 ± 0.0932 |
| RF | 0.6667 ± 0.0556 | 0.6649 ± 0.0443 | 0.6667 ± 0.0556 | 0.6572 ± 0.0444 |