Real-Time Methanol Detection in Distilled Spirits via BME688 MOX Sensor Array and Machine Learning

Methanol contamination in distilled beverages remains a serious public health risk, particularly in the context of illicit or poorly controlled production. Although confirmatory determination of methanol is routinely performed using laboratory-based techniques such as gas chromatography, spectrophotometry, and enzymatic assays, these approaches are often impractical for rapid screening due to infrastructure, expertise, time, and cost constraints. This study proposes a portable, fast and low-cost scanning framework that integrates the Bosch BME688 metal-oxide (MOX) sensor array with machine-learning models to detect methanol-related patterns in mixtures at varying ratios. Measurements in experiments were conducted on controlled ethanol-methanol mixtures and extended to real beverage matrices (Raki- and Whisky-based samples) to better reflect practical conditions where matrix effects may confound detection. After generating the raw dataset, the first preprocessing step involved denoising the multi-sensor, multi-step heater signals. The data were temporally aligned, a logarithmic transformation was applied, and standardisation was ensured. Models have been constructed using multiple heater profiles and a series of classification and regression algorithms to evaluate both categorical discrimination and quantitative prediction capabilities. The results demonstrate that MOX sensor responses, when paired with appropriate preprocessing and learning models, provide a viable basis for rapid, in-situ methanol screening in distilled beverages, supporting early warning and prioritisation prior to confirmatory laboratory analysis. Further studies are planned to expand the range of commercially consumed beverages and to evaluate the study's generalizability and limitations beyond controlled laboratory settings.