Adaptive filter for detection outlier data on electronic nose signal

Outliers are extreme cases that different from other observations. The existence of these outliers can damage data quality. The electronic nose (E-nose) is a tool for imitating the human nose work system that is cheap, accurate, and widely used to solve various scientific fields. Moreover, several feasibility studies have been conducted in recent years, such as detecting a mixture of beef and pork authenticity, diabetes patient detection, lung cancer, and respiratory infectious disease (SARS) detection based on the smell of armpit sweat. However, the inevitable challenge is detecting invalid data and preventing it from coming into the dataset. This paper proposes an adaptive filter using a deep neural network (DNN) and self-feature extraction to overcome the invalid data (outlier) presence from E-nose signal in case of SARS detection by armpit sweat odor. Our proposed method in the outlier detection task has a promising performance with 90.4% of average balanced accuracy (BA). These results indicate that the proposed DNN and self-feature extraction outperforms conventional methods such as Support Vector Machine (SVM), Naïve Bayes (NB), K-Nearest Neighbour (k−NN), and ensemble models like Random Forest (RF), XGBoost, also the Euclidean-z-score combination. The proposed system can be applied for real-time outlier detection on E-nose and improve the SARS detection performance.