Peak Selection in Fermentation Monitoring Using HS-GC-IMS: A Machine Learning and Topological Data Analysis Approach

Fermented foods have become increasingly popular in recent years, but fermentation processes are difficult to monitor because of their dynamic environment and the complexity of volatile organic compounds (VOCs). Traditional fermentation monitoring methods often neglect to detect VOCs, an important feature of the fermentation process. HS-GC-IMS (Headspace Gas Chromatography-Ion Mobility Spectrometry) provides a non-invasive and sensitive method for VOC detection, but generates large amounts of complex data that are difficult to analyze manually. To overcome this challenge, this study first optimizes the pre-processing steps such as region of interest (ROI) identification and hreshold selection, then combines Persistent Homology and Variable Importance in Projection (VIP) scores of PLS-DA, and finally succeeds in identifying the important peaks of VOCs generated during the fermentation process. This approach reduces manual intervention and improves the efficiency and accuracy of complex data processing. The contribution of this thesis is the combination of HS-GC-IMS with ML and TDA, which provides a new way for intelligent fermentation monitoring with potential application in the fermented food industry.