Why Can Multiple Imputations and How (MICE) Algorithm Work?

Multiple imputations compensate for missing data and produce multiple datasets by regression model and are considered the solver of the old problem of univariate imputation. The univariate imputes data only from a specific column where the data cell was missing. Multivariate imputation works simultaneously, with all variables in all columns, whether missing or observed. It has emerged as a principal method of solving missing data problems. All incomplete datasets analyzed before Multiple Imputation by Chained Equations (MICE) presented were misdiagnosed;results obtained were invalid and should not be countable to yield reasonable conclusions. This article will highlight why multiple imputations and how the MICE work with a particular focus on the cyber-security dataset. Removing missing data in any dataset and replacing it is imperative in analyzing the data and creating prediction models. Therefore, a good imputation technique should recover the missingness, which involves extracting the good features. However, the widely used univariate imputation method does not impute missingness reasonably if the values are too large and may thus lead to bias. Therefore, we aim to propose an alternative imputation method that is efficient and removes potential bias after removing the missingness.

应用轮廓变化信息的实验鼠行为识别

实验鼠行为分析数据是神经学、生理学、药物学等学科实验数据的重要部分。针对实验鼠缺少肢体运动信息的特点,提出一种实验鼠多行为分析方法。提取实验鼠轮廓的帧间变化信息,同时考虑变化信息与实验鼠本身的位置关系,对行为视频形成系列轮廓变化帧。在训练阶段,通过Pillar K-means聚类算法从系列帧中提取80个关键帧,并把每一个训练行为视频用对应关键帧频数的直方图表示。在测试阶段,测试视频用最近邻法确定每一帧对应的关键帧,形成相应的关键帧直方图,从而把分类问题变成一个直方图相似性问题,再应用卡方距离进行分类。实验结果表明,该方法对8种行为的分类准确率最高达到100%,最低达到95%。

基于遗传算法证券市场违规交易数据智能搜索

通过对我国证券市场中的内幕交易"老鼠仓"问题的研究,力图设计一个基于遗传算法的证券内幕交易异常数据自动检索系统,用来监控股票的违规操作.系统通过将遗传算法的快速搜索与异常处理判断模块结合,对出现异常波动的股票进行实时行情数据的跟踪、分析与调查,从中找出异常数据.之后再对这些异常数据进行比较和判断,定位该股是否具有违规的"老鼠仓"交易嫌疑,以此为证监部门判断某股是否有"老鼠仓"嫌疑提供依据.