<![CDATA[This research was conducted to develop a feature extraction system based on the dynamic characteristics of multisensor output data. The Quartz Crystal Microbalance (QCM) sensor was utilized as one type of multisensor that responds to changes in oscillation frequency for gas detection. The data used in this study were measurements of 6 types of mint species using the QCM sensor, taken at the Botanical Institute of Karlsruhe Institute of Technology (KIT) in Germany. The data underwent preprocessing to enhance the system's efficiency, followed by feature extraction using the Linear Discriminant Analysis (LDA) method. The feature extraction testing involved various data variations, resulting in a 58.33% reduction in the number of features from 12 to 5. Subsequently, classification testing was performed using three types of classifier models: k-Nearest Neighbors (k-NN), Support Vector Machine (SVM), and Decision Tree (DT). Four data variations were used for classification, which included frequency response data before and after LDA, as well as dynamic characteristic data extracted using Piecewise Linear Regression (PLR) before and after LDA. For the k-NN classification, the accuracies obtained were 96.05%, 98.40%, 93.75%, and 95% for each of the four data variations, with computation times of 0.402 s, 0.208 s, 0.027 s, and 0.023 s, respectively. For SVM classification, the accuracies obtained were 52.25%, 94.94%, 75.21%, and 87.92% for the four data variations, with computation times of 38.93 s, 3.749 s, 0.695 s, and 0.168 s, respectively. Lastly, for DT classification, the accuracies obtained were 97.95%, 96.09%, 86.25%, and 92.08% for the four data variations, with computation times of 0.647 s, 0.236 s, 0.066 s, and 0.023 s, respectively. Keywords: Quartz Crystal Microbalance, Feature Extraction, Linear Discriminant Analysis]]>
Copyrights © 2023
