Fuzzy c-Means Clustering-Based Novel Threshold Criteria for Outlier Detection in Electronic Nose

The presence of outliers deteriorates the overall performance of an electronic nose and, thus their detection and removal is critical to achieve the optimum discrimination ability among the subjected volatile organic compounds/gases. This article reports a semi supervised fuzzy c-means clustering based outlier detection method in the context of an electronic nose. A novel threshold criterion is developed using the fuzzy membership values and is used to evaluate whether the electronic nose response is an outlier or a true data sample. The method is tested on two experimentally generated electronic nose datasets using different types of sensor arrays. As the experimental datasets does not contain outliers, a mathematical outlier generation model is developed for outlier generation following the inherent properties of outliers. The proposed method effectively discriminates among the true data samples and the outliers for both the electronic nose datasets.