A Transparent Hybrid Machine Learning and Explainability Approach for Classifying Environmental Attitudes
Rukiye Uzun, Fuat Alkan, Irem Senyer Yapici
article (Uluslararası, Hakemli) | Yazar(lar): Rukiye Uzun, Fuat Alkan, Irem Senyer Yapici | Yıl: 2025 | Dergi/Kurum: IEEE ACCESS | İndeks: WOS.SCI | DOI: 10.1109/access.2025.3626005 | Anahtar Kelimeler: Accuracy, Radio frequency, Principal component analysis, Computational modeling, Education, Biological system modelin
Nowadays, the performance of machine learning (ML) models in data-driven decision-making processes largely depends on the quality of the data and the selection of informative features. In high-dimensional datasets, the presence of redundant or noisy attributes can hinder the learning process, lead to overfitting, and reduce model interpretability. Therefore, feature selection is considered a vital pre-processing step that not only reduces computational burden but also enhances model accuracy, generalizability, and transparency. Accurately classifying environmental attitudes is essential for addressing growing ecological challenges and fostering sustainable behaviors. In this context, the present study introduces an integrated ML and explainability methodology for the classification of environmental attitudes. The proposed approach integrates dimensionality reduction via Principal Component Analysis (PCA), feature ranking through Random Forest (RF), and explainable artificial intelligence (XAI) techniques to improve both prediction performance and model transparency. The novelty of this work lies in its innovative integration of these well-established techniques to provide transparent and interpretable results in the field of environmental behavioral science, specifically for the classification of environmental attitudes. Among the implemented models, the Multilayer Perceptron (MLP) supported by PCA and RF-based feature selection achieve the highest classification accuracy (98.7%), effectively capturing complex non-linear attitudinal patterns. To enhance interpretability, SHAP (SHapley Additive exPlanations) analysis has been employed, identifying the key survey items that contribute most to model decisions. The resulting interpretable outputs offer practical value for designing targeted sustainability education and policy initiatives, helping stakeholders better understand key drivers of environmental attitudes. Overall, the findings demonstrate that the integration of feature engineering and XAI provides a robust methodological foundation for environmental behaviour analysis, contributing to the development of data-driven and effective sustainability strategies.