This paper reports drastically improved accuracy of real-time gas detections by a scalable machine learning fusion system which utilizes encoder-decoder structures and a decision fusion model. Advancements as compared to the state-of-the-art works include: 1) increased gas sensing accuracy of a carbon dioxide sensor with a Mean Absolute Percentage Error (MAPE) of 2.97% in real-time tests; 2) drastically reduced sensor response/recovery time from ~9 mins to 2 mins; and 3) good proof-of-concept demonstrations for both generalization and robustness across different gas sensors. As such, this approach illustrates a new class of machine learning scheme to improve gas sensing results and potentially broad applications to other sensors.
Keywords: {Micromechanical devices;Accuracy;Systems architecture;Machine learning;Carbon dioxide;Real-time systems;Sensors;Time factors;Gas detectors;Standards;Gas Sensor;Real-Time Gas Detection;Machine Learning;Sensor Fusion},