Machine learning-based prediction model for high-value waste types to support the circular economy

Abstract

The management of high-value waste is a strategic challenge in supporting the implementation of a circular economy at the city level. This study aims to develop and compare machine learning-based prediction models to classify the economic value of waste into low, medium, and high categories based on the characteristics of transaction types and quantities. The dataset was obtained from waste bank transactions in Makassar City from January 2023 to June 2025. The research stages included data preprocessing, category labeling using the k-means and quantile-based classification algorithms, and then creating classification modeling using random forest and gradient boosting. The evaluation results show that both models perform very well with an accuracy rate above 92%, where gradient boosting consistently shows superior performance in terms of F1-score and the ability to reduce classification errors in high-value categories. Confusion matrix analysis shows that the boosting approach is more adaptive in handling class boundary distributions, while random forest tends to be more stable against data variations. These findings confirm that the ensemble learning approach is effective in supporting the classification of waste economic value based on historical transaction data.