Predicting willingness to pay for digital waste services using class-balanced machine learning

Abstract

As urban waste generation continues to increase, digital waste management applications offer a promising solution, yet their scalability is often constrained by low user adoption and uncertain willingness to pay (WTP). This study proposes a supervised machine learning (ML) framework to predict individual WTP for digital waste services using survey data from 400 residents of Gorontalo City, Indonesia, where fewer than 60% of households receive reliable waste collection. A random forest (RF) classifier with class-balancing and F2 score optimization was developed to prioritize recall of potential payers. The model achieved strong performance (F2 score: 0.81) and revealed occupation, app usability perception, and cost tolerance as top predictors. Feature importance analysis, correlation assessment, and boxplot visualization further clarified the behavioral and socio-economic drivers underlying payment decisions. By emphasizing recall over precision, the framework minimizes the risk of excluding willing participants and supports inclusive service planning. Beyond the local case study, the proposed approach applies to other cities with similar waste management constraints and emerging digital service ecosystems, offering a transferable decision-support tool for data-driven pricing, outreach, and policy design.