Metal oxide-based gas sensor array for VOCs determination in complex mixtures using machine learning

Detection of volatile organic compounds (VOCs) from the breath is becoming a viable route for the early detection of diseases non-invasively. This paper presents a sensor array of 3 component metal oxides that give maximal cross-sensitivity and can successfully use machine learning methods to identify four distinct VOCs in a mixture. The metal oxide sensor array comprises NiO-Au (ohmic), CuO-Au (Schottky), and ZnO-Au (Schottky) sensors made by the DC reactive sputtering method and having a film thickness of 80-100 nm. The NiO and CuO films have ultrafine particle sizes of < 50 nm and rough surface texture, while ZnO films consist of nanoscale platelets. This array was subjected to various VOC concentrations, including ethanol, acetone, toluene, and chloroform, one by one and in a pair/mix of gases. Thus, the response values show severe interference and departure from commonly observed power law behavior. The dataset obtained from individual gases and their mixtures were analyzed using multiple machine learning algorithms, such as Random Forest (RF), K-Nearest Neighbor (KNN), Decision Tree, Linear Regression, Logistic Regression, Naive Bayes, Linear Discriminant Analysis, Artificial Neural Network, and Support Vector Machine. KNN and RF have shown more than 99% accuracy in classifying different varying chemicals in the gas mixtures. In regression analysis, KNN has delivered the best results with an R² value of more than 0.99 and LOD of 0.012 ppm, 0.015 ppm, 0.014 ppm, and 0.025 ppm for predicting the concentrations of acetone, toluene, ethanol, and chloroform, respectively, in complex mixtures. Therefore, it is demonstrated that the array utilizing the provided algorithms can classify and predict the concentrations of the four gases simultaneously for disease diagnosis and treatment monitoring.