Possibility Data Analysis

本研究的目的是提出基于可能性判别概念的新的数据分析方法，将所提出的方法实现为计算机程序，并通过将其应用于实际问题来检验每种方法的能力.提出了一种由数值数据确定可能性回归模型系数的可能性分布的辨识方法。所提出的方法被实现为一个计算机程序，其性能进行了检查，应用到预制房屋的价格数据。提出了一种在多维模式空间中确定每类可能性分布的识别方法。识别的可能性分布线性映射到一个低维空间的特征向量。通过线性规划问题确定特征向量，将一类概率分布与其他类概率分布分离.提出了一种基于模糊神经网络的非线性可能性回归方法。提出了一种三角形模糊连接权值调整的学习算法.提出了一种基于模糊规则的回归方法，其中每个前件模糊集的隶属函数被视为一个可能性分布。将该方法与基于神经网络的方法进行了比较，并将其应用于米饭食味数据。

The following results were obtained by this research whose aims were to propose new data analysis methods based on the concept of possibility destribution, to implement the proposed methods as computer programs, and to examine the ability of each method by applying it to real-world problems.1. An identification method was proposed to determine a possibility distribution of the coefficients of a possibility regression model from numerical data. The proposed method was implemented as a computer program, and its performance was examined by the application to prefabricated house price data.2. An identification method was proposed to determine a possibility distribution of each class in a multi-dimensional pattern space. The identified possibility distribution was linearly mapped into a lower dimensional space by a characteristic vector. A linear propramming problem was formulated to determine this characteristic vector in order to separate the possibility distribution of one class from those of the other classes.3. A non-linear possibility regression method was proposed using fuzzy neural networks. A learning algorithm was derived to adjust triangular shape fuzzy connection weights.4. A fuzzy-rule-based regression method was proposed where the membership function of each antecedent fuzzy set was viewed as a possibility distribution. The proposed method was compared with a neural-network-based method by applying them to rice taste data.