基于代价敏感贝叶斯网络的烟叶感官质量评价

Tobacco Smoking Quality Evaluation Based on Cost-sensitive Bayesian Networks

贝叶斯网络在判别分类中具有很多优势,应用贝叶斯网络对烟叶感官质量进行预测和评价。一些烟叶质量指标的误分类代价不同,提出一种代价敏感贝叶斯网络。通过生成准则学习代价敏感贝叶斯网络的结构,进行代价敏感参数估计。应用代价敏感贝叶斯网络对一组烟叶进行感官质量预测和评价,结果表明了代价敏感贝叶斯网络在烟叶质量感官评价中的有效性。

Bayesian networks have many merits which are applied in data mine, tobacco smoking quality is predicted and evaluated using Bayesian networks. But some important smoking quality has unequal classification cost. Cost-sensitivity Bayesian network is proposed accordingly. Structure of the cost-sensitivity Bayesian networks is learned using generative criterion. Then parameters of the cost-sensitivity Bayesian networks are evaluated based on cost-sensitivity loss function. A Bayesian network is learned to form a tobacco smoking quality dataset. Experimental results show that cost-sensitivity Bayesian network is feasible to evaluate tobacco smoking quality.