2025 AIChE Annual Meeting

(588df) Integrating Machine Learning and Advanced Sensor Technologies for Predictive Flavor Profiling of Global Rums

Understanding the flavor profile of rum is essential for both consumers and producers, as it helps break down flavor attributes and assess their impact on the market. Currently, flavor evaluation relies on trained panelists, which introduces subjectivity and limits reproducibility. This study proposes an approach based on advanced chemical techniques and machine learning to support these panelists. To achieve this, rum samples from various regions worldwide were analyzed using Gas Chromatography-Mass Spectrometry (HS-SPME-GC-MS) along with flavor tools such as the electronic tongue (E-tongue) and electronic nose (E-nose), identifying key volatile compounds in their aromatic profile. The collected data was analyzed using statistical and machine learning techniques. Principal Component Analysis (PCA) reduced dimensionality and revealed key patterns in chemical and flavor composition. Unsupervised methods like k-means identified natural groupings of rums based on molecular behavior, raw material, and geographical origin. A PERMANOVA analysis was applied to assess the statistical significance of these groupings regarding their region of origin. Additionally, predictive models were developed to identify molecules based on aroma and flavor profiles. The Random Forest model achieved a Hamming accuracy of over 80% in molecular prediction, outperforming Support Vector Machines (SVM) and neural networks. Furthermore, variable importance analysis determined which flavor attributes had the greatest influence on rum classification based on aromatic profile. The combination of chemical analysis, electronic organoleptic techniques, and machine learning algorithms not only improved the understanding of rum’s flavor profile but also provided deeper insights into its key determinant factors. These findings offer innovative tools to enhance rum characterization and development within the industry.