基于机器学习对泌尿类疾病标志物气体识别模式研究

Research on Gas Recognition Pattern of Urinary Disease Markers Based on Machine Learning

摘要: 泌尿类疾病，例如膀胱癌和前列腺癌，对全球健康构成严重威胁。本研究针对泌尿类疾病标志性VOC气体(甲苯，乙苯，异丙醇，戊醛)，通过选取电子鼻采集的多传感器信号与这四类气体的关联特征，采用常规气体传感器特征，建立四分类VOC分类预测模型。采用主成分分析(PCA)对样本点降维，分别使用三种分类算法：K-邻近(KNN)，支持向量机(SVM)和随机森林(RF)进行分类预测，三者准确率分别达到了88%，85%和91%。最后使用Stacking集成方式，分别对KNN和SVM，KNN和RF，SVM和RF进行两两集成，集成后的准确率有明显提升，其中效果最佳的集成方式是SVM和RF，其准确率达到了97%。研究表明stacking集成的SVM和RF模型成功地预测四种标志物VOC，为泌尿类关键疾病的早期筛查和无创检测打下坚实基础。

Abstract: Urinary diseases, such as bladder cancer and prostate cancer, pose a serious threat to global health. This study focuses on the landmark VOC gases (toluene, ethylbenzene, isopropanol, and glutaraldehyde) of urinary diseases. By selecting the correlation characteristics between the multi-sensor signals collected by the electronic nose and these four gases, and using conventional gas sensor features, a four class VOC classification prediction model is established. Principal Component Analysis (PCA) was used to reduce the dimensionality of sample points, and three classification algorithms were used: K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Random Forest (RF) for classification prediction, with accuracy rates of 88%, 85%, and 91%, respectively. Finally, using the Stacking integration method, KNN and SVM, KNN and RF, SVM and RF were integrated pairwise, and the accuracy was significantly improved after integration. The best integration method was SVM and RF, with an accuracy of 97%. Research has shown that the SVM and RF models integrated with stacking have successfully predicted four biomarkers of VOC, laying a solid foundation for early screening and non-invasive detection of key urological diseases.