基于多类支持向量机的化学物质生态危害分类研究

Classification of Chemicals by Ecological Hazard Using Multi-Class Support Vector Machines

应用多类支持向量机(M-SVMs)方法研究了化学物质生态危害程度的分类,以提高分类的准确性和效率。对采集到的61种环境优先污染物的环境行为和生物毒性方面的7项指标进行相关性分析,建立了M-SVMs分类模型并对数据集进行10折交叉验证以评价模型的分类能力,运用所建模型对7种化学物质的生态危害进行预测。结果表明,去除与鱼毒有信息重叠的溞毒指标,选取鱼毒、藻毒、降解性、蓄积性、分配系数和吸附系数6项指标用于构建M-SVMs分类模型;M-SVMs模型识别率较高,交叉验证平均分类正确率达86.89%;对7种化学物质生态危害的预测结果与实际情况基本相符。

Classification of chemicals by ecological hazard was studied with the multi-class support vector machines （M-SVMs） to improve accuracy and efficiency of the classification. A total of 61 environmental priority pollutants that had already been collected were analyzed for correlation between 7 indexes in the aspects of environmental behavior and biotoxicity. On such a basis a M-SVMs classification model was established and 10-fold cross validation of its dataset was conducted to evaluate classification ability of the model. Then the model was applied to predict ecological hazard of 7 chemicals. Results show that the index of daphnia toxicity, overlapping some of the information of the index of fish toxicity, was excluded from the 7 indexes. So only 6, i. e. as fish toxicity, alga toxicity, biodegradability, bioconcentration, distributivity and adsorbability were selected in building the M-SVMs classification model. The M-SVMs model was quite high in identification rate and cross-validation indicated that its mean classification accuracy reached up to 86.89%, and its prediction of the 7 chemicals in ecological hazard basically tallies with the actual situation.