数据挖掘中并行离散化数据准备优化

Parallel Discretization of Data Preparation Optimization in Data Mining

摘要:

在海量数据挖掘中，针对元数据的离散化数据准备处理能有效提高数据挖掘效率[1]。本文提出了一种并行比较并获得最优离散化的数据准备算法（AOA），针对不同数据集，先进行数据集的特性检测以获得数据集分布特性，按照分布特性进行数据集的异常值检测和剔除，并行完成与分布特性适配的离散化方法处理，通过比较不同离散化方法的熵、方差指数、稳定性参数的最小欧氏距离[2]，根据三个参数自动化比选，获得最优离散化的预处理成果。仿真表明，对不同样本数据库进行关联规则挖掘结果中，比较四种固定的离散化数据预处理方法，在使用AOA数据准备算法并行比选出最优的离散化来数据预处理后，在不同最小支持度阈值情况下，挖掘得到关联规则数都更少，因此效率得到提高。

Abstract:

In data mining, the discretization of data can improve the efficiency of data mining effectively. In this paper, we propose a data preprocessing algorithm (AOA) to obtain the optimal discretization using parallel comparison. For different data sets, we first perform the feature detection of the data set to obtain the distribution characteristics of the data set. Then the outliers of the data set are detected according to the distribution characteristics. IN addition, the discretization results are obtained by comparing the minimum Euclidean distance of the entropy, the variance index and the stability parameter of the different discretization methods. In simulation experiment, we compare the AOA with four typical data discretization methods in different databases by running the association rule mining algorithm on the discretization data obtained using AOA and other four methods, respectively. The results show that, under different minimum support thresholds, the number of association rules extracted from the discretization data obtained using AOA is the least, indicating higher efficiency of AOA.