A PCA–machine learning framework for understanding household food waste: evidence from young urban consumers in Albania

Abstract

This study explores the predictors of household food waste among young consumers (N = 414) in Albania, a country undergoing a transition from traditional to urbanized food systems. Using Principal Component Analysis (PCA), we identified eight dietary patterns, three waste patterns, three categories of reasons for food waste, and one dimension of dietary association. These components were analyzed alongside demographic characteristics through Random Forest Regression (RFR) and Artificial Neural Networks (ANN) to evaluate their predictive capacity. The results show that food waste is systematically linked to dietary regimes: perishable fresh foods are wasted due to storage and planning deficits, while protein and convenience-based diets drive waste through over-purchasing and portioning errors. Forest Regression (RFR) models consistently outperformed Artificial Neural Networks (ANNs) in predictive accuracy, with higher R2 values (0.47–0.62 vs. 0.15–0.37) and lower error rates, demonstrating the strength of combining PCA with ML techniques. The findings highlight the behavioral pathways behind waste and provide a novel approach to modeling sustainability challenges in transitioning food systems.