STATISTICAL METHODS FOR RECURSIVELY PARTITIONED TREES

描述(改编自申请者摘要)：该拟议项目有 三个主要目标。目标1是制定改进的战略，以 适合更准确的分类和回归树(即CART)模型。 目标2是制定一个正式的框架，以便对以下方面进行统计推断 树模型。目标3是开发和分发公共领域软件 这将允许应用程序数据分析人员实现我们在 前两个目标。为了实现这些目标，我们将整合 统计和计算机器学习方法。我们相信我们的工作 可以通过将 统计数据在制定模型选择客观标准方面的优势 并提供评估和量化不确定性的框架 与模型相关联，并利用机器学习的优势进行拟合 从模型到大型而复杂的数据集。

DESCRIPTION (Adapted from the Applicant's Abstract): This proposed project has three primary objectives. Objective 1 is to develop improved strategies for fitting more accurate classification and regression tree (i.e., CART) models. Objective 2 is to develop a formal framework to allow statistical inference on tree models. Objective 3 is to develop and distribute public-domain software that will allow applied data analysts to implement the methods we develop in the first two objectives. To meet these objectives we will integrate statistical and computational machine learning approaches. We believe our work can have a significant impact in biomedical data analysis by combining the strengths of statistics for developing objective criteria for model selection and for providing a framework for assessing and quantifying uncertainty associated with a model, with the strengths of machine learning for fitting models to large and complex datasets.