Machine learning-driven integration of GC-MS and sensory panel data for aroma prediction in food systems

Understanding food aroma remains challenging due to the complex interactions of volatiles and the subjectivity of sensory evaluation. The integration of gas chromatography-mass spectrometry (GC-MS), sensory analysis, and machine learning (ML) offers a powerful framework to model aroma perception. This review synthesizes advances in ML-driven aroma prediction between 2020 and 2025, encompassing more than 60 peer-reviewed studies that link GC-MS fingerprints to sensory descriptors across diverse food systems. Reported models achieved prediction accuracies typically ranging from 70 % to 99 %, with ensemble and deep learning methods frequently outperforming linear baselines. We examine key ML architectures-from ensemble methods to deep learning-and their ability to identify aroma-driving compounds and predict sensory profiles. Case studies from coffee, wine, dairy, fermented products, and plant-based alternatives are discussed. Emphasis is placed on interpretability, predictive accuracy, and analytical rigor. Methodological challenges such as sensory variability, overfitting, and generalizability are critically evaluated. Finally, we explore emerging directions including data fusion, sensory-guided design, and real-time aroma profiling, positioning ML as a catalyst for predictive and scalable flavor science.