Optimizing threshold using pearson correlation for selecting features of electronic nose signals

This research studied the use of sensors in an electronic-nose (e-nose) system to select the most efficient features to detect three classes of data signals. The data signals were collected from sensors that were assembled specifically for recognition of meat types. The development of e-noses has been the subject of many researches. However, researchers tend to install multiple sensors without optimizing their utilization, leading to higher costs. This research proposes three significant contributions to optimize the number of sensors to be used in e-noses, i.e. (i) tuning the threshold of the correlation coefficient; (ii) obtaining significant features by determining the optimal threshold of independent features; (iii) optimizing efficiency by considering the significant features and the accuracy of the results. The proposed efficiency method suggested that the number of sensors in the system could be reduced from 10 to 5. The optimal threshold suggested that 12 out of 20 features were correlated to each other. The highest accuracy obtained in this research was 92% for three classes of meats.