基于先验决策模型的缺失数据谱系树构建方法研究

发布时间：2018-08-31 09:05

【摘要】：谱系树用来表示物种之间的亲缘关系,是生命科学研究的重要内容,而形态学数据是构建古生物谱系树的重要基础。但是,古生物形态学数据常有缺失信息,造成传统谱系树构建算法失效,针对该问题本文提出了基于先验决策模型的缺失数据谱系树构建方法。具体研究内容如下:(1)针对含有缺失数据谱系树构建的问题,提出了基于先验决策模型的谱系树构建方法。首先,利用较完整数据的物种集构建初始谱系树;然后,运用属性约简原理得到决策点,进而建立先验决策模型;最后,根据模型确定缺失数据比例较高的物种在初始谱系树中的位置,通过物种嫁接完成谱系树的构建。实验结果表明,当单个物种缺失数据比例大于10%时,相比最大简约法在平均准确率方面高出10%左右。(2)决策点中属性决策组集合的生成是一个属性组合优化问题,提出了运用遗传算法的属性决策组集合构建方法。该方法结合待约简属性个数与组合类别划分,提出了特有的编码方式、相应的适应度函数和交叉变异方式。进一步提高了缺失数据物种在谱系树中嫁接位置的准确性和稳定性。实验结果表明,本方法所构建的谱系树在物种平均准确率方面相比简单启发式方法最多高出3.4%。(3)为了消除因多次嫁接而产生的并系子树,提出了一种基于权值拟合层次聚类的并系子树预测方法。首先利用最小二乘法对初始谱系树进行属性权值的拟合,进而对并系物种属性进行加权,最后运用有权值层次聚类的方法构建子树。实验结果表明,本方法能够对并系子树进行预测,最终消除了谱系树中的并系子树。
[Abstract]:Pedigree is an important part of life science research, and morphological data is an important basis for constructing paleontology pedigree. However, paleontology data often have missing information, resulting in the failure of traditional pedigree construction algorithm. To solve this problem, this paper proposes a priori decision-making model based on the lack of information. The specific research contents are as follows: (1) To solve the problem of constructing missing data pedigrees, a method of constructing pedigrees based on a prior decision model is proposed. Firstly, the original pedigrees are constructed by using a more complete set of species; then, the decision points are obtained by using attribute reduction principle, and then a prior decision is made. Finally, the position of the species with high missing data proportion in the initial pedigree tree is determined according to the model, and the pedigree tree is constructed by species grafting. The experimental results show that the average accuracy of the attribute decision-making set is about 10% higher than that of the maximum reduction method when the missing data proportion of a single species is greater than 10%. Combining the number of attributes to be reduced with the classification of combinatorial classes, a special encoding method, the corresponding fitness function and the crossover mutation method are proposed, which further improves the grafting of missing data species in the pedigree tree. The experimental results show that the average accuracy of species is up to 3.4% higher than that of the simple heuristic method. (3) In order to eliminate the parallel subtree caused by multiple grafting, a new method based on weight fitting hierarchical clustering is proposed to predict the parallel subtree. The least squares method is used to fit the attribute weights of the initial pedigree tree, and then weights the attributes of the merged species. Finally, a subtree is constructed by using the weighted hierarchical clustering method. The experimental results show that this method can predict the merged subtrees and eliminate the merged subtrees in the pedigree tree.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：Q11;TP18

【相似文献】