A Multivariable Regression-Based Freshness Estimation Approach for Food Waste Reduction and Supply Chain Management

Summary Food sustainability and safety are among the critical concerns of 21st-century development. However, the difficulty of making effective food freshness estimates and predictions poses significant limitations to food sustainability. This study proposes a multivariable regression and sensor data fusion approach for assessing and predicting food freshness to demonstrate the feasibility of accurately estimating freshness in perishable foods, which can support waste reduction and supply chain improvement. This approach addresses the limitations of traditional techniques in terms of efficiency and cost. Machine learning algorithms were employed to analyze multisensor data (temperature, humidity, gas concentrations, etc.) and construct freshness evaluation models. The experiments focused on two representative scenarios: (1) packaged oranges in a virtual cold chain, where regression trees and feedforward neural networks achieved the best performance (root mean square error (RMSE) is as low as 0.0068); and (2) radio frequency identification (RFID) sensors-monitored stored hams, with the nonlinear auto-regressive exogenous (NARX) model demonstrating superior accuracy in time-series data (RMSE = 0.0007). The results validate the effectiveness of the proposed method in improving freshness prediction.