Several methods have been used to detect infectious respiratory diseases, for example, by taking samples from blood, saliva, and phlegm. Although these methods generated high accuracy, they raised more problems that increased the risk of spreading and required more time to detect. Therefore, an accurate, quick, and low-cost device is required to help detect infectious respiratory diseases. This study proposes a new approach for detecting infectious respiratory diseases using an electronic nose (E-nose) through sweat samples from the human axilla. E-nose became safer by taking samples through the axillary because infectious respiratory diseases are not transmitted through sweat. This study proposes two new feature extraction techniques called stable value and highest slope. This study also proposes a stacked Deep Neural Network (DNN) for effective infectious respiratory disease detection. In the proposed stacked DNN, five fine-tuned DNN models obtained from hyperparameter tuning are stacked then the output of each DNN model became the input of the meta-model in the form of a fully connected layer. The proposed feature extraction method outperformed the existing feature extraction and was able to separate data between classes better. Furthermore, the proposed stacked DNN model generated an accuracy of 0.934 in the testing data, outperforming DNN single models and other state-of-the-art machine learning algorithms.
- Deep learning
- Electronic nose
- Feature extraction
- Infectious respiratory disease