Predicting the Shelf Life of Cut Tuberose (Polianthes tuberosa) Using Support Vector Regression: Model Development, Benchmarking, and Practical Insights

The study developed and evaluated models to predict the shelf life of cut tuberose (Polianthes tuberosa) from routinely measured postharvest indicators. An experimental-style dataset (n=220) was assembled during 2023–24, comprising browning index, physiological loss in weight (PLW), moisture content, flower opening index, freshness index, microbial load ( CFU ), total phenols (mg GAE ), storage temperature, relative humidity, and pretreatment (Control, Sucrose+Germicide, Pulsing+STS). Data were standardized and one-hot encoded; support vector regression (SVR, RBF kernel) was tuned via five-fold cross-validation and benchmarked against ridge regression and a tuned random forest. Ridge exhibited the highest generalization (test , MAE=0.771 d, RMSE=0.991 d; mean CV- ), followed by SVR (test , MAE=0.922 d, RMSE=1.148 d; mean CV- ). A compact tuning sweep modestly improved random forest (test , MAE=1.039 d, RMSE=1.328 d; mean CV- ). Permutation-based importance indicated storage temperature and microbial load as principal determinants, with hydration metrics (PLW, moisture), freshness, and pretreatment contributing additional signal. It was concluded that day-level forecasts were feasible from a parsimonious measurement suite, and that temperature control, sanitation, hydration, and carbohydrate–germicide pretreatments were the most effective levers to extend marketable life.