Fresh produce quality is a critical determinant of consumer preference and market value, with high-quality tomatoes often fetching premium prices. Traditional quality assessment methods, though effective, are typically labor-intensive, destructive, and impractical for real-time monitoring. In this study, we explore the potential of low-cost ($0.99 per unit) metal oxide semiconductor (MOS) gas sensors—MQ series (MQ2, MQ5, MQ6, MQ7, MQ9, and MQ135)—as a rapid, non-destructive, and cost-efficient tool for distinguishing high-quality tomatoes based on their cultivation in nitrogen-rich or nitrogen-deficient soils. Organic tomatoes, harvested from plants exposed to relatively high (High-N) and relatively low (Low-N) levels of bioavailable nitrogen (N) were enclosed in airtight containers equipped with MQ sensors to capture their volatile organic compound (VOC) emissions over 48 hours. Three replicates were used for each N level. Complementary laboratory-based quality assessments measured fresh weight, soluble sugar content, titratable acidity, pH, firmness, color space (Lab*), antioxidant capacity (DPPH assay), and total phenolic content. Gas chromatography-mass spectrometry (GC-MS) was utilized for VOC profiling. The resulting sensor and analytical data were preprocessed and normalized, followed by training 193 machine learning models with principal component analysis (PCA) at a 95% variance threshold in MATLAB. Significant differences were observed between High-N and Low-N treatments across several quality parameters. High-N tomatoes exhibited a 25.32% increase in average fruit fresh weight (p = 0.002), whereas Low-N tomatoes had 18.80% higher firmness (p = 0.020). Low-N tomatoes showed a 27.09% increase in antioxidant capacity (p = 0.0001), a possible indication of an adaptive response to N deficiency. Whereas VOC analysis revealed higher concentrations of octanoic acid, nonanoic acid, and 2-methyl-1-propanol in High-N tomatoes, with increases of 142.67%, 191.46%, and 37.72%, respectively, compared to Low-N tomatoes (p = 0.007, p = 0.020, p = 0.040). Sensor performance analysis demonstrated that MQ9 and MQ5 sensors were the most effective in differentiating between the two nitrogen treatments, with ensemble learning, neural networks, and support vector machines achieving 100% classification accuracy, followed by MQ135 and MQ2. Feature reduction criterion identified a minimal yet highly discriminative subset—including MQ9 sensor responses, octanoic acid, 4-heptanone, nonanoic acid, 1-penten-3-ol, 2-methyl-1-propanol, limonene, 3-methylbutanoic acid, 2-heptanone, fresh weight, and DPPH values—yielding classification accuracies of 97.06% during training and 89.29% in testing with a tri-layer neural network model. These findings underscore the potential of low-cost MOS gas sensors, particularly MQ9, as a viable, non-destructive alternative for real-time quality assessment of tomatoes.
