A Hybrid ML–FDEA Framework for Efficient and Uncertainty-Aware Biomass Supply Chain Optimization

Abstract

This paper introduces a new hybrid model, combining both Machine Learning (ML) and Fuzzy Data Envelopment Analysis (FDEA), to optimize the biomass supply chain management in bioethanol production systems. The study concentrates on the effective transportation of excess sugarcane bagasse at the sugar mills to bioethanol factories in Western Maharashtra region. To simulate the dynamics of a real-world supply chain, a detailed multi-year dataset (2017-2025) was built based on production, transportation, and cost-related parameters gained through official reports, and then preprocessed and engineered features according to the systematic standards. The FDEA model proposed assesses the efficiency of plant-mill transportation routes by considering various inputs and outputs under uncertainty, with the efficiency scores being between 0 and 1. Through the analysis, it is found that there is a high variability in route performance as the values of efficiency lie between 0.072 and 0.465, illustrating the effect that transportation distance and cost have on the efficiency of the supply chain. In order to have better scalability and predictability, machine learning models were trained to predict FDEA efficiency scores. Ensemble learning methods such as AdaBoost, Random Forest and Decision Tree were found to perform strongly on predicting with R2 of 0.950, 0.943 and 0.947 respectively and custom deep learning as implemented in PyTorch and TensorFlow were found to outperform with R2 of 0.975 and 0.978 respectively. The findings confirm that the hybrid ML-FDEA model is an effective method to incorporate efficiency assessment, uncertainty modelling and predictive analytics, such that the biomass supply chain optimization framework offers a strong decision support system. The framework has a better accuracy, scalability and viable applicability than the current methods. The research makes a contribution to sustainable bioenergy systems, whereby biomass logistics can be optimized cost-effectively and by data to optimize biomass logistics in the face of real-world uncertainties.