MetaNChemo: A meta-heuristic neural-based framework for chemometric analysis

Gas sensors are present in almost every environmental monitoring application. They can be realized using different materials and production techniques while exploiting different physical/chemical processes to detect one or more target gas. Historically, their research and development are mainly driven by the materials science community. However, the seamless combination of artificial intelligence techniques with gas sensors in cyber–physical systems is recently attracting the interest of various computer science communities. In this paper, we propose MetaNChemo, a multidisciplinary end-to-end framework to build a chemometric system from the very production of chemoresistive sensors, through the data sampling and pre-processing phases necessary to calibrate such sensors, to the identification and the assessment of the most suitable artificial intelligence models able to supports the detection of the concentration of a target gas in the air. Without loss of generality, we focus our attention on carbon monoxide (CO) as our target gas. However, materials scientists may take full advantage of MetaNChemo, since its data-driven approach can accommodate very specific requirements of those environmental monitoring applications that have to properly detect one or more target gas. By resorting to state-of-the-art meta-heuristic techniques, we identify and train tiny neural networks (20-50 weights) able to achieve F1-scores greater than 0.95 over real environmental test conditions.