Evaluating the effect of aroma diffusion on E-Nose sensor performance and classification accuracy of pineapple ripeness levels

Background The Electronic Nose (E-Nose) system detects and analyzes odors using gas sensors to identify Volatile Organic Compounds (VOCs) emitted by an object. This study used the E-Nose system to classify the ripeness levels of pineapples based on the detected odor. One factor influencing classification accuracy is the distance between the sensor and the object. The sensor generates 47 features. Therefore, this research aims to explore the effect of sensor distance on the classification accuracy of pineapple ripeness levels using classification algorithms, namely Support Vector Machine (SVM), Adaptive Boosting (AdaBoost), and Gradient Boosting. Methods This study involved three pineapple ripeness classes—Ripe, Half-Ripe, and Unripe—each with 25 samples. Each sample weighed 50 grams and was cut into uniformly sized cubes. Data were collected using nine different Ming Qin (MQ)-type Metal Oxide Semiconductor (MOS) gas sensors at two measurement distances: 15 and 65 cm. Each sample generated 150 rows of Comma Separated Value (CSV) data. The collected data were processed using machine learning algorithms to classify the ripeness levels of the pineapples. The algorithms—SVM, AdaBoost, and Gradient Boosting—were calibrated using GridSearchCV to obtain optimal parameters. Results Based on the experimental results, at a 15 cm measurement distance, all algorithms demonstrated higher accuracy compared to the 65 cm distance. SVM and AdaBoost achieved 90% accuracy at 15 cm, while Gradient Boosting recorded 86.67%. However, at 65 cm, SVM’s accuracy dropped to 83.33%, AdaBoost to 76.67%, and Gradient Boosting recorded 80% accuracy. Discussion These results indicate that sensor distance significantly impacts classification accuracy, with closer distances yielding more accurate results. Additionally, analysis using the confusion matrix for each algorithm showed that SVM delivered the best performance in predicting pineapple ripeness classes at both measurement distances.