A Combined Ontology and Machine Learning Framework for Predicting Asbestos in Buildings

Asbestos remains a serious public health concern, particularly in older buildings, despite its ban in France in 1997. While both ontology and machine learning techniques have been used independently in various domains, their combination offers a novel approach for enhancing the detection and classification of asbestos-containing materials. This research introduces an innovative framework that integrates an ontology-based semantic model with decision tree algorithms and Semantic Web Rule Language (SWRL) rules to address the challenges of accurate asbestos identification. The use of ontology allows us to incorporate domain knowledge and semantic relationships, improving model interpretability and classification precision. In this study, we compared three approaches: a decision tree model without ontology, SWRL rules applied to a numerical ontology (where categorical variables were encoded), and SWRL rules applied to a categorical ontology (preserving the original categorical values). Our results show that the categorical ontology, when combined with SWRL rules, significantly outperforms the other models in classifying buildings as containing or not containing asbestos. However, we also observed that missing data in the dataset negatively impacts the performance of the ontology-based approach, highlighting the limitations of semantic models in data-incomplete environments. This work demonstrates the potential of combining ontology and machine learning to enhance asbestos detection while identifying the challenges that arise from incomplete data.