Gas leakage detection is imperative in various sectors, including chemical industries, coal mines, and household applications. The escalating number of accidents in coal mines, chemical industries, and homes underscores the urgency of swift and accurate gas detection methods. This research focuses on developing advanced systems that promptly identify gas types to prevent harm to human lives and the environment. This paper addresses the challenges of gas leakage detection and classification in diverse environments, such as industrial, residential, and mining scenarios. The proposed ExAIRFC-GSDC model integrates machine learning algorithms, particularly a Random Forest Classifier, with explainable artificial intelligence (XAI) techniques to enhance interpretability. This study employs a dataset comprising gas sensor measurements that encompassing gasses, such as Liquid Petroleum Gas (LPG), Compressed Natural Gas (CNG), Methane, Propane, and others. Various machine learning classifiers, including K-Nearest Neighbors, Decision Tree, Support Vector Machines, XGBoost, and others, are compared with ExAIRFC-GSDC for gas detection. The model demonstrates superior performance, achieving an accuracy rate of 98.67%. Incorporating SHAP and LIME explanations enhances the model's interpretability, providing insights into the contributions of individual sensors. Statistical analysis confirms the significant differences in sensor readings across different gas types. ExAIRFC-GSDC is a robust and explainable solution for accurate gas detection and classification in complex environments.
Read full abstract