Machine Learning–Based Prediction of Beef Tomato Exocarp Firmness Using Physicochemical and Mineral Nutrient Profiles
JOURNAL OF FOOD SCIENCE, cilt.91, sa.10, ss.1-25, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 91 Sayı: 10
- Basım Tarihi: 2026
- Doi Numarası: 10.1111/1750-3841.71558
- Dergi Adı: JOURNAL OF FOOD SCIENCE
- Derginin Tarandığı İndeksler: Applied Science & Technology Source, Food Science & Technology Abstracts, Academic Search Ultimate (EBSCO), Natural Science Collection (ProQuest), Biomedical Reference Collection: Corporate Edition (EBSCO), Engineering Source (EBSCO), Health Research Premium Collection (ProQuest), Scopus, Science Citation Index Expanded (SCI-EXPANDED), BIOSIS, Chemical Abstracts Core, EMBASE, Environment Index, INSPEC, MEDLINE, DIALNET
- Sayfa Sayıları: ss.1-25
- Akdeniz Üniversitesi Adresli: Evet
Özet
Tomato exocarp firmness (TEF) is an important mechanical quality parameter that directly affects postharvest handling, transportation, storage stability, and industrial processing performance. Conventional firmness determination methods are generally destructive, time-consuming, and unsuitable for rapid food quality monitoring systems. This study investigated the relationship between physicochemical composition and TEF and developed machine learning-based models for data-driven firmness prediction. A total of 163 beef tomato samples were analyzed using 11 physicochemical and mineral nutrient variables, including total soluble solids (TSS), dry matter, titratable acidity, lycopene (LYC), vitamin C, calcium, potassium, magnesium, manganese, sodium, and phosphorus. Deep learning (DL), artificial neural networks (ANNs), and multiple linear regression (MLR) models were developed and evaluated using coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE). Correlation analysis demonstrated strong positive relationships between exocarp firmness and vitamin C (r = 0.677), LYC (r = 0.668), and manganese (r = 0.622), whereas phosphorus (r = −0.660) and TSSs (r = −0.635) showed strong negative correlations. Among the predictive approaches, the DL model achieved the best performance with an R2 value of 0.709, RMSE of 0.158, and MAE of 0.132. ANNs and MLR produced lower predictive accuracy with R2 values of 0.673 and 0.605, respectively. The findings demonstrate that physicochemical and mineral nutrient profiles can support data-driven prediction of TEF. However, because the predictors were obtained through laboratory analyses, further validation using independently collected data and nondestructive sensing technologies is required before practical industrial application.