Machine Learning‐Based Prediction of Post‐Harvest Losses in Egyptian Strawberry Exports: Cold Chain Analytics, Shelf‐Life Estimation, and Export Rejection Risk Modeling

Post‐harvest losses are the most economically damaging phase of quality deterioration in the Egyptian strawberry export corridor. Egypt is the world's leading exporter of individually quick‐frozen strawberries and a top‐three supplier of fresh strawberries globally, yet substantial losses arise from cold chain failures, suboptimal packaging, and logistical inefficiencies. The objectives of this study are (1) to develop the first ML‐based post‐harvest loss prediction framework calibrated to Egyptian strawberry export conditions; (2) to benchmark four model architectures (Random Forest, eXtreme gradient boosting [XGBoost], support vector regression/support vector classification [SVC], and long short‐term memory [LSTM]) for simultaneous shelf‐life estimation, export rejection classification, and loss severity categorization; and (3) to apply SHapley Additive exPlanations (SHAP) explainability analysis to identify the dominant controllable cold chain predictors and translate model outputs into actionable guidance for cold chain operators, packinghouse managers, and export logistics coordinators. The framework was trained on a physics‐calibrated synthetic dataset of 800 Egyptian export lots (30 input features, 10 target variables). An ensemble stacking model achieved shelf‐life prediction R ² = 0.934 (root mean square error = 0.41 days), export rejection area under the curve‐receiver operating characteristic = 0.961 (F1 = 0.862), and loss category quadratic weighted kappa = 0.887. SHAP analysis identified pre‐cooling delay, storage temperature, temperature deviation events, and transport duration as the four dominant predictors.