Scholarlink

<p>Predicting firefighter interventions is challenging due to the high dimensionality and complexity of the data. This study introduces a hybrid feature selection framework that combines ontology-based reasoning with machine learning (ML) techniques to improve predictive accuracy and model interpretability. Three ML algorithms—XGBoost, LightGBM, and LSTM—were applied using two feature selection strategies: a traditional ML-based approach and a hybrid method integrating ontology-driven centrality metrics (degree, closeness, betweenness). A domain-specific ontology was developed to capture key factors like environmental and temporal variables, enhancing feature selection for more relevant and interpretable inputs. The hybrid approach consistently outperformed the ML-only method across all models, achieving <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10115_2025_2497_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="21" /> </InlineMediaObject> <EquationSource Format="TEX">\(R^2\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> scores of 0.976 (XGBoost), 0.964 (LSTM), and 0.975 (LightGBM), compared to slightly lower scores for ML-only approaches. These findings demonstrate the potential of combining ontology-based reasoning with ML to address high-dimensional data challenges in predictive tasks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A hybrid ontology-based feature selection framework for enhancing predictive accuracy in regression models

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Sarah Ayad,
  • Roxane Elias Mallouhy,
  • Christophe Guyeux

摘要

Predicting firefighter interventions is challenging due to the high dimensionality and complexity of the data. This study introduces a hybrid feature selection framework that combines ontology-based reasoning with machine learning (ML) techniques to improve predictive accuracy and model interpretability. Three ML algorithms—XGBoost, LightGBM, and LSTM—were applied using two feature selection strategies: a traditional ML-based approach and a hybrid method integrating ontology-driven centrality metrics (degree, closeness, betweenness). A domain-specific ontology was developed to capture key factors like environmental and temporal variables, enhancing feature selection for more relevant and interpretable inputs. The hybrid approach consistently outperformed the ML-only method across all models, achieving \(R^2\) R 2 scores of 0.976 (XGBoost), 0.964 (LSTM), and 0.975 (LightGBM), compared to slightly lower scores for ML-only approaches. These findings demonstrate the potential of combining ontology-based reasoning with ML to address high-dimensional data challenges in predictive tasks.