Comparison of different feature reduction methods in the improvement of gas diagnosis of a temperature modulated resistive gas sensor

The present study aims to analyze dynamic responses of a temperature modulated resistive gas sensor with the emphasis on the comparison of different feature reduction methods. For this purpose, four selected feature reduction methods consist of Linear Discriminant Analysis (LDA), Principal Component Analysis (PCA), Generalized-LDA (GDA) and Kernel-PCA (KPCA) are applied and compared. The sensor selected for the experiment is a tin oxide based sensor, FIS commercial type. A staircase voltage with the step length of 40 s and voltage range of 1-5 V constitutes the input of the sensor. Sensor system was modeled by ARMAX linear model. The effects of induced gases were recorded as parameter vectors in the data obtained by the model. After applying the methods of feature reductions, the performance of gas separation was compared. It was found out that LDA and GDA yielded the best data classification.