A robust machine learning framework for predicting the higher heating value of poultry litter using proximate analysis

Date

2026-08-01

Author

Eren, Beytullah
Uzun, Suleyman
Özdemir, Serkan

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

1526
views

0
downloads

Cite This

Accurate estimation of the higher heating value (HHV) of poultry litter is critical for its valorization as a renewable energy feedstock. However, the heterogeneity of poultry litter and data scarcity pose challenges for predictive modeling. In this study, a robust machine learning framework was developed to predict HHV using dry-basis proximate analysis parameters (volatile matter, fixed carbon, and ash). Unlike previous studies that rely on standard datasets, this work introduces a systematic comparison of four modeling scenarios to isolate the effects of data augmentation and hyperparameter optimization. Gaussian noise-based augmentation was formulated to expand the training space, while GridSearch cross-validation optimized model parameters. Five algorithms (KNN, RF, Extra Trees, LGBM, XGBoost) were evaluated and compared against a Multiple Linear Regression (MLR) baseline. Results indicated that models on the original dataset suffered from overfitting (Test R-2 0.89, RMSE = 0.80 MJ/kg), outperforming the MLR baseline (R-2 = 0.64) and other ML models. Feature importance and SHAP interaction analysis confirmed that ash content is the dominant inhibitor of energy density, aligning with thermochemical principles. The proposed framework offers a rapid, low-cost alternative to expensive elemental analysis for industrial energy assessment.

URI

https://hdl.handle.net/11511/118867

Collections

Graduate School of Informatics, Article