A hybrid machine learning model for predicting agricultural production costs: Integrating economic sensitivity analysis and environmental factors in Egypt.

Abstract

Accurate prediction of agricultural production costs is crucial for sustainable development in Egypt, where productivity is highly sensitive to fluctuating economic and environmental conditions. This study introduces a hybrid machine learning model that provides more accurate and actionable cost predictions than existing models by uniquely integrating machine learning techniques with time-series forecasting of key economic variables and robust economic sensitivity analysis, all within a framework that also considers environmental factors. Using a 20-year dataset (2004-2023) on tomato production, we find that the Support Vector Machine (SVM) model outperforms Random Forest (RF) and Decision Tree (DT) models across three growing seasons (Summer, Nile, Winter), achieving a 2 % improvement in R². Key cost drivers include human labor wages, irrigation water costs, and minimum temperature. Time-series forecasting reveals projected increases in inflation and fuel prices, underscoring the need for proactive policy interventions. Sensitivity analysis identifies fuel prices and inflation as the most influential economic factors, with varying impacts across seasons. This integrated approach offers actionable policy recommendations to enhance food security, economic resilience, and environmental sustainability in Egypt's agricultural sector, with broader implications for Africa.