從氣體感測器的響應曲線中萃取特徵可以獲得氣體和感測器材料之間的反應的信息，並用於後續的氣體模型。但是，冗餘的特徵可能會降低氣體分類的正確性，冗餘特徵產生的原因可能為氣體感測陣列因為使用時間和製程的關係，而使得訊號產生飄移和老化，進而導致再現性的降低，特徵選擇演算法可以解決這個問題。
特徵選擇演算法可大致分成過濾法和包裝法，但這兩個方法都存在著一些問題。現在使用在電子鼻的過濾法沒有針對再現性這個特性去討論，而包裝法則是很依賴分類器，一旦選定用甚麼分類器評估特徵，後續分類時的分類器也一定要相同，通用性低。
本研究提出了最小距離內點概率（MDIP）特徵選擇演算法。透過考慮特徵的分離性和再現性並配合評估策略，MDIP可以有效地篩選掉冗餘特徵並提供有代表性的特徵子集以達到更好的分類正確性，且不限定於特定分類器。本研究使用分別紀錄製程變異和感測器飄移問題的數據來驗證MDIP的可行性與效能。實驗結果證明，採用MDIP特徵選擇演算法，兩個數據集的平均分類正確率分別高出使用所有特徵下的狀況46.1％和37.5％，比起Linear SVMRFE則是分別高出約10.1%和4.7%。

To obtain meaningful information from data, feature extraction methods are applied to the gas sensor responses of electronic nose systems. However, redundant features may cause variation and diminish the accuracy of gas classification. The sources of variation could be the lack of reproducibility during manufacture and process variation between the same type of sensors. Fortunately, feature selection provides a solution.
Feature selection is a widely used technique, and its methods can be divided into two categories: filters and wrappers, but both of them have some problems. Filters used in the electronic nose failed to consider reproducibility, and wrappers need to use the specific classifier, lacking of generality.
In this paper, a minimum distance inlier probability (MDIP) feature selection (FS) method is proposed. By incorporating the intrinsic properties of features and ranking strategy, MDIP can efficiently eliminate redundant features and provide better classification accuracy and is not tied to a specific classifier. The performance of the method was validated on two open-access datasets that provide information for system variation and sensor drift problems, respectively. Experimental results revealed that the average classification accuracy for the two datasets was higher than using all features by 46.1% and 37.5%, respectively, with the MDIP method. The results also showed that the average classification accuracy for the two datasets was higher than Linear SVM-RFE by 10.1% and 4.7%, respectively, with the MDIP method.

[1] T. Eklöv, P. Mårtensson, I. Lundström, Enhanced selectivity of MOSFET gas sensors by systematical analysis of transient parameters, Analytica Chimica Acta (1997), vol. 353(2-3), pp. 291-300.
[2] S. Zhang, C. Xie, M. Hu, H. Li, Z. Bai, D. Zeng, An entire feature extraction method of metal oxide gas sensors, Sensors and Actuators B: Chemical (2008), vol. 132, pp. 81–89.
[3] R. Gutierrez-Osuna, A. Gutierrez-Galvez, N. Powar, Transient response analysis for temperature-modulated chemoresistors, Sensors and Actuators B: Chemical (2003), vol. 93, pp. 57–66
[4] E. Martinelli, C. Falconi, A. D’Amico, C. Di Natale, Feature extraction of chemical sensors in phase space, Sensors and Actuators B: Chemical (2003), vol. 95, pp. 132–139.
[5] X.R. Wang, J.T. Lizier, A.Z. Berna, F.G. Bravo, S.C. Trowell, Human breath-print identification by E-nose, using information-theoretic feature selection prior to classification, Sensors and Actuators B: Chemical (2015), vol. 217, pp. 165–174.
[6] Y. Saeys, I. Inza, P. arrañaga, A review of feature selection techniques in bioinformatics, bioinformatics (2007), vol. 23(19), pp. 2507-2517.
[7] M.K. Muezzinoglu, A. Vergara, R. Huerta, M.I. Rabinovich, A sensor conditioning principle for odor identification, Sensors and Actuators B: Chemical (2010), vol. 146, pp. 472–476.
[8] K. J. Johnson, S. L. Rose-Pehrsson, Sensor array design for complex sensing tasks, Annual Review of Analytical Chemistry (2015), vol. 8, pp. 287-310.
[9] C. Deng, K. Lv, D. Shi, B. Yang, S. Yu, Z. He, J. Yan, Enhancing the discrimination ability of a gas sensor array based on a novel feature selection and fusion framework, Sensors (2018), vol. 18(6), pp. 1909.
[10] T. Nowotny, A. Z. Berna, R. Binions, S. Trowell, Optimal feature selection for classifying a large set of chemicals using metal oxide sensors, Sensors and Actuators B: Chemical (2013), vol. 187, pp. 471-480.
[11] I. Guyon, J. Weston, S. Barnhill, V. Vapnik, Gene selection for cancer classification using support vector machines, Machine learning (2002), vol. 46(1-3), pp. 389-422.
[12] M. Peris, L. Escuder-Gilabert, A 21st century technique for food control: Electronic noses, Analytica chimica acta (2009), vol. 638(1), pp. 1-15.
[13] B. Kateb, M. A. Ryan, M. L. Homer, L. M. Lara, Y. Yin, K. Higa, M. Y. Chen, Sniffing out cancer using the JPL electronic nose: A pilot study of a novel approach to detection and differentiation of brain cancer, NeuroImage (2009), vol. 47, pp. T5-T9.
[14] J.W. Gardner, P.N. Bartlett, A brief history of electronic nose, Sensors and Actuators B: Chemical (1994), vol. 18, pp. 210-211.
[15] A. Campagnoli, F. Cheli, C. Polidori, M. Zaninelli, O. Zecca, G. Savoini, L. Pinotti, V. Dell’Orto, Use of the Electronic Nose as a Screening Tool for the Recognition of Durum Wheat Naturally Contaminated by Deoxynivalenol: A Preliminary Approach, Sensors (2011), vol. 11, pp. 4899–4916.
[16] J. Gebicki, B. Szulczy´nski, Discrimination of selected fungi species based on their odour profile using prototypes of electronic nose instruments, Measurement (2018), vol. 116, pp. 307–313.
[17] B. Scholkopf, A. J. Smola, Learning with kernels: support vector machines, regularization, optimization, and beyond (2001), MIT press.
[18] K. G. Derpanis, Overview of the RANSAC Algorithm, Image Rochester NY (2010), 4(1), pp. 2-3.
[19] S. Choi, T. Kim, W. Yu, Performance evaluation of RANSAC family, Journal of Computer Vision (1997), vol. 24(3), pp. 271-300.
[20] J. Fonollosa, L. Fernandez, A. Gutiérrez-Gálvez, R. Huerta, S. Marco, Calibration transfer and drift counteraction in chemical sensor arrays using Direct Standardization, Sensors and Actuators B: Chemical (2016), vol. 236, pp. 1044-1053.
[21] A. Vergara, S. Vembu, T. Ayhan, M. A. Ryan, M. L. Homer, R. Huerta, Chemical gas sensor drift compensation using classifier ensembles, Sensors and Actuators B: Chemical (2012), vol. 166, pp. 320-329.