Scholarlink

<p>Fire accidents have displaced people from their homes and hindered community development, underscoring the critical importance of early smoke detection to safeguard lives and property. To address this, the present study proposes intelligent smoke detection using machine learning models, including Support Vector Classification (SVC), Gaussian Process Classifier (GPC), and Light Gradient Boosting Classifier (LGBC). Furthermore, novel hybrid models are introduced by integrating these classifiers with the Manta Ray Foraging Optimization (MRFO) algorithm to enhance predictive accuracy. Among these, the LGMR model—an optimized version of LGBC using MRFO—demonstrated superior performance. It achieved the highest F1-scores of 0.850 and 0.970 for false and true alarm classifications, respectively, along with a precision of 0.990 and recall of 0.950 for true alarms, indicating its strong ability to accurately distinguish between actual fire events and false positives. The dataset used in this study comprises numerically preprocessed sensor readings, and the models were evaluated using accuracy, precision, recall, and F1-score metrics. The findings affirm that the LGMR hybrid model significantly improves smoke detection reliability and holds considerable promise for real-world fire safety applications.</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

Design and implementation of a comprehensive classification framework for distinguishing true and false outputs in smoke sensor systems

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Sainan Wang,
  • Wei Wei

摘要

Fire accidents have displaced people from their homes and hindered community development, underscoring the critical importance of early smoke detection to safeguard lives and property. To address this, the present study proposes intelligent smoke detection using machine learning models, including Support Vector Classification (SVC), Gaussian Process Classifier (GPC), and Light Gradient Boosting Classifier (LGBC). Furthermore, novel hybrid models are introduced by integrating these classifiers with the Manta Ray Foraging Optimization (MRFO) algorithm to enhance predictive accuracy. Among these, the LGMR model—an optimized version of LGBC using MRFO—demonstrated superior performance. It achieved the highest F1-scores of 0.850 and 0.970 for false and true alarm classifications, respectively, along with a precision of 0.990 and recall of 0.950 for true alarms, indicating its strong ability to accurately distinguish between actual fire events and false positives. The dataset used in this study comprises numerically preprocessed sensor readings, and the models were evaluated using accuracy, precision, recall, and F1-score metrics. The findings affirm that the LGMR hybrid model significantly improves smoke detection reliability and holds considerable promise for real-world fire safety applications.