Enhanced selectivity of platinum-modified tungsten oxide gas sensor through multivariate feature extraction and machine learning algorithms.

The limited selectivity of metal oxide semiconductor (MOS) gas sensors poses a significant challenge in accurately identifying volatile organic compounds (VOCs) within industrial environments. Here, platinum-modified tungsten oxide (Pt/WO) composite was successfully prepared through in-situ reduction, which not only possesses superior gas-sensing performance towards ppm-level triethylamine but also achieves robust humidity resistance and long-term stability. Benefiting from the catalytic sensitization of noble metal, the as-fabricated Pt/WO sensor exhibits improved sensitivity towards triethylamine as compared with the pristine tungsten oxide (WO) sensor. To minish low-frequency noise and promote selectivity, the sensor response curves were fitted and processed with discrete wavelet transform (DWT) for 4-level decomposition. Accompanied with multivariate feature extraction, an artificial neural network (ANN) algorithm was developed to categorize and identify multiple VOCs including triethylamine, ammonia, and isopropanol, enabling a decision boundary with 95.2 % accuracy. Moreover, the prediction of unknown gas concentration was successfully achieved by linear regression model after training a series of as-known concentrations. This work not only offers a rational solution to design high-performance gas sensors but also provides an intelligent strategy to identify gas concentration under multiple VOCs.