An In2O3 Nanotubes based Gas Sensor Array combined with Machine Learning Algorithms for Trimethylamine Detection

This work focuses on developing an electronic nose system with machine learning algorithm for detection of trimethylamine (TMA). Pure and Ga-doped In2O3 nanotubes are synthesized by a simple electrospinning method, and four kinds of gas sensors (pristine, 1% Ga, 10% Ga, and 20% Ga-doped In2O3) are fabricated to form a sensor array. Results show that the sensor array can classify TMA effectively from interference gases (xylene, ethanol, hydrogen sulfide) by a support vector machine (SVM) algorithm. Several algorithms, including radial basis function neural network (RBFNN), back propagation neural network (BPNN) and principal component analysis combined with linear regression (PCA-LR), are used to predict the concentration level of each gas. For TMA gas, the trained algorithms can predict its concentration with average relative errors of 1.22% for RBFNN, 2.5% for BPNN and 13.34% for PCA-LR. Furthermore, the binary mixtures of TMA and ethanol are measured and used to train the above algorithms, and the lowest average relative error of 1.74% is achieved in the case of RBFNN algorithm.