Evaluating the storage stability of tomato paste using chemical preservatives: Assessing its quality attributes and predictive modeling with machine learning

This study investigated the effects of chemical-based preservation on the physicochemical properties, bioactive compounds, and microbial stability of tomato paste stored at 4-6°C for 8 months. Four samples were prepared, S1 (100 mg/kg sodium metabisulfite+500 mg/kg sodium benzoate+10000 mg/kg citric acid), S2 (100 mg/kg potassium metabisulfite+500 mg/kg sodium benzoate+10000 mg/kg citric acid), S3 (100 mg/kg sodium metabisulfite+100 mg/kg potassium metabisulfite+500 mg/kg sodium benzoate+10000 mg/kg citric acid), and control without additives. To predict the degradation trends of bioactive compounds, four machine learning models were utilized. Ridge Regression was used for linear regularization and Support Vector Machine for kernel-based mapping; Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), were implemented to capture complex, non-linear trends. During storage, total sugar and acidity increased, while total soluble solids, lycopene, vitamin C, total phenolic content, and DPPH• scavenging activity declined across all treatments. Treated samples exhibited no detectable total plate count, yeast, mold, E. coli , or Salmonella , whereas the control sample spoiled within one month. In machine learning, RF and XGBoost showed superior performance where XGBoost achieved low RMSE for most bioactive compounds. Therefore, chemical treatment can preserve tomato paste for 8 months and RF, XGBoost models can be utilized for bioactive compound prediction.