Machine Learning Approaches to Environmental Cost Allocation in Manufacturing Firms

Abstract

This paper introduces a novel framework for environmental cost allocation in manufacturing firms by integrating machine learning techniques with activity-based costing principles. Traditional environmental accounting methods often rely on simplistic
allocation bases that fail to capture the complex, non-linear relationships between production activities and their environmental impacts. Our research addresses this gap
by proposing a hybrid methodology that combines unsupervised learning for activity clustering and supervised learning for impact prediction. We develop a two-stage
model: first, using self-organizing maps to identify homogeneous activity clusters based
on multi-dimensional environmental drivers; second, employing gradient boosting machines to predict environmental costs for each cluster. The methodology was validated
using data from three manufacturing firms with diverse production processes. Results
demonstrate that our approach reduces allocation errors by 42% compared to traditional volume-based methods and by 28% compared to conventional activity-based
costing. Furthermore, the model reveals previously unrecognized cost drivers, including machine idle time patterns and raw material quality variations, which significantly
influence environmental costs but are typically overlooked in standard accounting systems. The framework provides manufacturing managers with more accurate environmental cost information, enabling better decision-making for sustainable production.
This research contributes to both environmental accounting and machine learning applications by demonstrating how advanced analytics can transform traditional cost allocation practices, offering a more nuanced understanding of environmental cost causality
in complex manufacturing environments.