Fusion of Feature Ranking Methods for an Effective Intrusion Detection System

Expanding internet-connected services has increased cyberattacks,many of which have grave and disastrous repercussions.An Intrusion Detection System(IDS)plays an essential role in network security since it helps to protect the network from vulnerabilities and attacks.Although extensive research was reported in IDS,detecting novel intrusions with optimal features and reducing false alarm rates are still challenging.Therefore,we developed a novel fusion-based feature importance method to reduce the high dimensional feature space,which helps to identify attacks accurately with less false alarm rate.Initially,to improve training data quality,various preprocessing techniques are utilized.The Adaptive Synthetic oversampling technique generates synthetic samples for minority classes.In the proposed fusion-based feature importance,we use different approaches from the filter,wrapper,and embedded methods like mutual information,random forest importance,permutation importance,Shapley Additive exPlanations(SHAP)-based feature importance,and statistical feature importance methods like the difference of mean and median and standard deviation to rank each feature according to its rank.Then by simple plurality voting,the most optimal features are retrieved.Then the optimal features are fed to various models like Extra Tree(ET),Logistic Regression(LR),Support vector Machine(SVM),Decision Tree(DT),and Extreme Gradient Boosting Machine(XGBM).Then the hyperparameters of classification models are tuned with Halving Random Search cross-validation to enhance the performance.The experiments were carried out on the original imbalanced data and balanced data.The outcomes demonstrate that the balanced data scenario knocked out the imbalanced data.Finally,the experimental analysis proved that our proposed fusionbased feature importance performed well with XGBM giving an accuracy of 99.86%,99.68%,and 92.4%,with 9,7 and 8 features by training time of 1.5,4.5 and 5.5 s on Network Security Laboratory-Knowledge Discovery in Databases(NSL-KDD),Canadian Institute for Cybersecurity(CIC-IDS 2017),and UNSW-NB15,datasets respectively.In addition,the suggested technique has been examined and contrasted with the state of art methods on three datasets.

FAST FEATURE RANKING AND ITS APPLICATION TO FACE RECOGNITION

A fast feature ranking algorithm for classification in the presence of high dimensionahty and small sample size is proposed. The basic idea is that the important features force the data points of the same class to maintain their intrinsic neighbor relations, whereas neighboring points of different classes are no longer to stick to one an- other. Applying this assumption, an optimization problem weighting each feature is derived. The algorithm does not involve the dense matrix eigen-decomposition which can be computationally expensive in time. Extensive exper- iments are conducted to validate the significance of selected features using the Yale, Extended YaleB and PIE data- sets. The thorough evaluation shows that, using one-nearest neighbor classifier, the recognition rates using 100-- 500 leading features selected by the algorithm distinctively outperform those with features selected by the baseline feature selection algorithms, while using support vector machine features selected by the algorithm show less prominent improvement. Moreover, the experiments demonstrate that the proposed algorithm is particularly effi- cient for multi-class face recognition problem.

Feature Selection for Image Classification Based on a New Ranking Criterion

In this paper, a feature selection method combining the reliefF and SVM-RFE algorithm is proposed. This algorithm integrates the weight vector from the reliefF into SVM-RFE method. In this method, the reliefF filters out many noisy features in the first stage. Then the new ranking criterion based on SVM-RFE method is applied to obtain the final feature subset. The SVM classifier is used to evaluate the final image classification accuracy. Experimental results show that our proposed relief- SVM-RFE algorithm can achieve significant improvements for feature selection in image classification.

Expert ranking method based on ListNet with multiple features

The quality of expert ranking directly affects the expert retrieval precision.According to the characteristics of the expert entity,an expert ranking model based on the list with multiple features was proposed.Firstly,multiple features was selected through the analysis of expert pages;secondly,in order to learn parameters through gradient descent and construct expert ranking model,all features were integrated into ListNet ranking model;finally,expert ranking contrast experiment will be performed using the trained model.The experimental results show that the proposed method has a good effect,and the value of NDCG@1 increased14.2%comparing with the pairwise method with expert ranking.

Development of Data Mining Models Based on Features Ranks Voting (FRV)

Data size plays a significant role in the design and the performance of data mining models.A good feature selection algorithm reduces the problems of big data size and noise due to data redundancy.Features selection algorithms aim at selecting the best features and eliminating unnecessary ones,which in turn simplifies the structure of the data mining model as well as increases its performance.This paper introduces a robust features selection algorithm,named Features Ranking Voting Algorithm FRV.It merges the benefits of the different features selection algorithms to specify the features ranks in the dataset correctly and robustly;based on the feature ranks and voting algorithm.The FRV comprises of three different proposed techniques to select the minimum best feature set,the forward voting technique to select the best high ranks features,the backward voting technique,which drops the low ranks features(low importance feature),and the third technique merges the outputs from the forward and backward techniques to maximize the robustness of the selected features set.Different data mining models were built using obtained selected features sets from applying the proposed FVR on different datasets;to evaluate the success behavior of the proposed FRV.The high performance of these data mining models reflects the success of the proposed FRV algorithm.The FRV performance is compared with other features selection algorithms.It successes to develop data mining models for the Hungarian CAD dataset with Acc.of 96.8%,and with Acc.of 96%for the Z-Alizadeh Sani CAD dataset compared with 83.94%and 92.56%respectively in[48].

Performance prediction for Grid workflow activities based on features-ranked RBF network

Accurate performance prediction of Grid workflow activities can help Grid schedulers map activitiesto appropriate Grid sites.This paper describes an approach based on features-ranked RBF neural networkto predict the performance of Grid workflow activities.Experimental results for two kinds of real worldGrid workflow activities are presented to show effectiveness of our approach.

基于多源知识和Ranking SVM的中文微博命名实体链接

命名实体是文本中承载信息的重要单元,正确分析存在歧义的命名实体对文本的理解起着关键性作用。该文提出基于多源知识和Ranking SVM的中文微博命名实体链接,结合同义词词典、百科资源等知识产生初始候选实体集合,同时从文本中抽取多种组合特征,利用Ranking SVM对候选实体集合进行排序,从而得到目标实体。在NLP&CC20141中文微博实体链接评测数据集上进行了实验,获得了89.40%的平均准确率,与NLP&CC2014中文微博实体链接评测取得最好成绩的系统相比,本文的系统具有一定的优势。

基于多特征和Ranking SVM的微博新闻自动摘要研究

提出了面向微博应用的新闻文本自动摘要研究方法.利用互信息对新闻文本中词语和句子之间的语义特征进行计算,根据其关联度对句子进行主题划分,赋予主题句较高的权重,同时从文本中抽取多种组合特征,利用Ranking SVM对句子进行排序,从而得到自动摘要.在NLP&CC2015面向微博中文新闻自动摘要评测数据集上进行对比实验,取得了良好效果,证明该方法的有效性.

Feature Selection Based on Difference and Similitude in Data Mining

Feature selection is the pretreatment of data mining. Heuristic search algorithms are often used for this subject. Many heuristic search algorithms are based on discernibility matrices, which only consider the difference in information system. Because the similar characteristics are not revealed in discernibility matrix, the result may not be the simplest rules. Although differencesimilitude（DS） methods take both of the difference and the similitude into account, the existing search strategy will cause some important features to be ignored. An improved DS based algorithm is proposed to solve this problem in this paper. An attribute rank function, which considers both of the difference and similitude in feature selection, is defined in the improved algorithm. Experiments show that it is an effective algorithm, especially for large-scale databases. The time complexity of the algorithm is O（｜ C ｜^2｜U ｜^2）.

基于6种机器学习算法的早发性卵巢功能不全影响因素分析

目的:通过机器学习算法对早发性卵巢功能不全(POI)的影响因素进行特征排序,找出对POI影响较大的因素。方法:先制定纳入和剔除标准,选取因月经不调就诊的500例患者,根据中医证型进行年龄和职业差异性分析。再通过逻辑回归、支持向量机、决策树、随机森林、极端梯度提升和K-最近邻6种机器学习算法对患者进行POI预测分类,根据算法求得的马修斯相关系数和AUC进行预测精准度比较。通过随机森林中的准确度和基尼不纯度下降对POI影响因素进行特征排序,结合逐步剔除法得到对POI影响程度排序前五的特征因素。结果:随机森林的算法在马修斯相关系数、准确率和AUC中均获得了最大值,分别为0.399、0.717和0.908。POI的影响因素有子宫或盆腔手术史、受教育程度、年龄、减肥史和吸烟史,这些因素的Borda计数得分依次为手术史(2.446)、受教育程度(2.924)、年龄(4.060)、减肥史(5.303)、吸烟史(6.429)。结论:随机森林的性能在预测POI患者中优于其他5种算法,当患者的数据信息不足时,医生可先通过这5个特征因素的指标对月经不调患者进行初步干预。

高阶煤吸附孔结构特征及其对甲烷吸附能力的影响

孔隙结构对煤层吸附甲烷的能力有显著影响,但目前对高阶煤吸附孔结构特征及其对甲烷吸附能力的影响研究较少。以贵州兴安煤业有限公司糯东煤矿高阶煤样为研究对象,采用低温N2吸附和低温CO_(2)吸附试验,结合分形理论研究了高阶煤吸附孔的孔隙结构特征,并通过高压等温甲烷吸附试验,分析了煤储层物性、孔隙结构特征和分形维数对甲烷吸附能力的影响。结果表明:①高阶煤储层孔隙形态较为单一,多数为两端开放的平行板孔和狭缝型孔,微孔在煤的孔隙结构中占主导地位,其孔体积和孔比表面积占比均大于98%,为气体的富集提供了空间。②以不同孔径段的孔体积占比为权重计算高阶煤孔隙的综合分形维数,微孔分形维数在综合分形维数中占主导地位;煤样孔隙结构具有明显的分形特征,孔隙非均质性较强。③Langmuir模型能很好地描述高阶煤的吸附行为,煤储层物性、孔隙结构和分形维数对甲烷吸附能力影响显著,Langmuir体积与最大镜质体反射率、镜质组含量、灰分含量和水分含量呈线性正相关关系,与惰质组含量呈线性负相关关系;Langmuir体积与吸附孔的孔比表面积和孔体积均呈线性正相关关系,Langmuir体积与分形维数呈弱线性关系。研究结果可为黔西南地区高阶煤层气勘探开发及煤矿瓦斯灾害防治提供理论指导。

基于三重混合采样和集成学习的潜在高价值旅客发现

考虑潜在高价值旅客特有的数据高度不平衡、旅客特征和价值类别弱相关等问题,提出一种基于三重混合采样和集成学习的潜在高价值旅客发现模型。采用RFM(Recency Frequency Monetary)方法标注旅客类别;使用三重混合采样对不平衡旅客数据集进行重采样;使用融合特征选择算法遴选旅客特征;使用梯度提升决策树作为分类器,构建旅客价值预测模型,识别潜在高价值旅客。在PNR数据集上的实验结果表明,与基准算法相比,该模型能取得更好的AUC值和F1值,可以较好地识别潜在高价值旅客。

技术评审专家遴选方法在颠覆性技术专家预判平台上的应用

评审专家遴选是技术评审中的关键环节。鉴于颠覆性技术专家预判平台预判系统对时效性和智能型的要求,专家遴选对预判结果具有决定性影响。通过学术专长匹配和专业遴选来选择符合要求的专家,可以降低成本,提高推荐效率与准确度,完成颠覆性技术的预测任务。基于学术网络表示学习的方法既可以避免大量特征工程,又可以方便不同类型的特征进行融合。利用异质网络表示学习方法和标签排序的学术专长画像方法构建专家库,并使用融合专家综合评价指标特征的匹配方法对待预判的颠覆性技术和专家专长进行匹配,为专家遴选提供一份专业背景匹配的候选专家列表。这种方法在Academic Social Network数据集上进行模拟实验。实验结果表明,这种方法能提升项目评审专家学术专长匹配,在加入综合指标特征后,专家的综合指标特征能有效地反馈到实验结果中,从而提高评审系统的时效性和智能性。

基于改进TF-IDF融合二进制灰狼优化的短文本分类

为了提高特殊类型短文本分类准确度和降低特征维度,提出了基于改进TF-IDF方法融合二进制灰狼优化的短文本分类。为了提高特征向量文本权重计算准确度,提出了点赞排列因子,并融合了文本特征集中度,对附有点赞数的特殊类型文本进行权重计算,设计改进了TF-IDF-RANK方法对特征进行加权;同时,基于初选特征向量,设计优化了二进制灰狼优化算法(BGWO)搜寻最优特征子集,引入衰减系数向量和多优解迭代机制,提高灰狼搜寻性能。结果表明,该方法有效地提升了权重准确率,更好地表征初选特征向量,增强特征选择时寻找全局最优解的能力,进而提高短文本的分类效果。通过LABIC和抖音开放平台数据集测试,综合指标F1值分别提高了14.76%和14.02%,验证了该方法对于特殊类型文本分类的有效性。

面向无监督特征提取的结构化图嵌入

特征提取是处理高维数据的最有效工具之一。然而,当前特征提取方法存在两个问题:一是它们没有同时捕捉数据的局部和全局结构;二是构建的图脱离数据的聚类数,没有与聚类相同的连通分量。为了解决这些问题,提出了面向无监督特征提取的结构化图嵌入方法(structured graph embedding, SGE)。通过构建数据表征的K近邻和使用最小二乘回归,SGE能够同时保持数据的局部与全局相关结构。而且,SGE对表征图的拉普拉斯矩阵施加秩约束,使构建的最优图具有与c个聚类一致的c个连通分量,从而能揭示数据的聚类结构。因此,SGE能够找到更有判别力的投影。在多个真实数据集的实验表明,SGE优于其他主流降维方法。特别是在PIE数据集上,SGE的聚类精度比LRPP_GRR的聚类精度高出18.7%。这些结果表明SGE方法可以有效降低数据维数。

基于子空间字典低秩表示的流形投影学习

低秩表示(Low-Rank Representation,LRR)能够将每个数据点表示为若干个基的线性组合,是一种获取样本底层低维结构的方法。然而,大多数LRR方法使用原始数据集作为字典,这不能揭示数据的真实分割。本文提出了基于子空间字典低秩表示的流形投影学习:该方法学习最优子空间作为LRR问题的字典,而不是使用原始数据集;利用基数最少的方案,低秩表示矩阵能很好地恢复原始数据;通过对投影矩阵施加行稀疏约束,该方法不仅可以选择鉴别性特征并忽略冗余特征,而且使子空间学习具有很好的解释性。此外,通过引入流形结构保持约束,使得样本的原始表示和距离信息在投影下保持不变。在多个真实世界数据集上的实验结果表明,该方法优于最近提出的一些相关方法。

莫斯科国际大学排名体系MosIUR研究

本文以莫斯科国际大学排名MosIUR为研究对象,阐释其构建动因和方法论原则,通过对比分析与统计学方法,揭示其区别于其他高校排名体系THE、QS、U.S. News、ARWU的差异性特征。MosIUR以大学的三大使命为出发点,采用客观的定量标准,衡量高校在“教育教学”“科学研究”和“社会贡献”三方面的能力。“社会贡献”是该排名体系独设的评价指标。MosIUR的建立为世界大学排名提供了新的参照体系,它与其他排名形成一定的互补。本研究有助于全面了解MosIUR排名体系以及当下俄罗斯对高等教育发展的定位与考量,也为思考我国高等教育建设以及科技革命即将带给世界大学排名体系的影响提供参考和借鉴。

低秩自适应微调的一阶段红外目标跟踪

由于缺乏大规模的红外跟踪训练数据集,现有的红外跟踪方法大都利用在大规模可见光数据上预训练的模型,然后在小规模的红外数据上进行完全微调。然而,当预训练模型的参数规模迅速增大时,完全微调需要的内存和时间成本也急剧增长,这限制了低资源用户在大型模型上进行研究和应用。为解决该问题,提出一种参数、内存和时间高效自适应的红外目标跟踪算法。首先,通过Transformer的自注意力机制对模板和搜索区域图像进行联合特征提取和关系建模,获取与目标关联度更强的特征表示;其次,利用低秩自适应矩阵的侧网络将可训练参数从主干网络中进行解耦,以减少需要训练更新的参数规模;最后,设计一种轻量级空间特征增强模块,增强特征对目标和背景的判别能力。提出方法的训练参数,内存和时间分别仅占完全微调方法的0.04%、39.6%和66.2%,性能却超越了完全微调。在3个标准红外跟踪数据集LSOTB-TIR120,LSOTB-TIR100和PTB-TIR上的实验对比结果和消融实验证明了提出的方法是有效的。提出的方法在LSOTB-TIR120数据集上成功率为73.7%,精度为86.0%,归一化精度为78.5%;LSOTB-TIR100数据集上成功率为71.6%,精度为83.9%,归一化精度为76.1%;在PTB-TIR数据集上成功率为69.0%,精度为84.9%,均取得了当前最先进的跟踪性能。

Detecting Malicious Uniform Resource Locators Using an Applied Intelligence Framework

The potential of text analytics is revealed by Machine Learning(ML)and Natural Language Processing(NLP)techniques.In this paper,we propose an NLP framework that is applied to multiple datasets to detect malicious Uniform Resource Locators(URLs).Three categories of features,both ML and Deep Learning(DL)algorithms and a ranking schema are included in the proposed framework.We apply frequency and prediction-based embeddings,such as hash vectorizer,Term Frequency-Inverse Dense Frequency(TF-IDF)and predictors,word to vector-word2vec(continuous bag of words,skip-gram)from Google,to extract features from text.Further,we apply more state-of-the-art methods to create vectorized features,such as GloVe.Additionally,feature engineering that is specific to URL structure is deployed to detect scams and other threats.For framework assessment,four ranking indicators are weighted:computational time and performance as accuracy,F1 score and type error II.For the computational time,we propose a new metric-Feature Building Time(FBT)as the cutting-edge feature builders(like doc2vec or GloVe)require more time.By applying the proposed assessment step,the skip-gram algorithm of word2vec surpasses other feature builders in performance.Additionally,eXtreme Gradient Boost(XGB)outperforms other classifiers.With this setup,we attain an accuracy of 99.5%and an F1 score of 0.99.

具有混合策略的樽海鞘群特征选择算法

近年来,随着计算机和数据库技术的快速发展,大规模数据集迅速增长,利用特征选择技术来筛选信息量大的特征已经变得非常重要。本文提出了一种具有混合策略的樽海鞘群特征选择算法(salp swarm feature selection algorithm with hybrid strategy,HS-SSA)。首先,本文生成一张基于互信息的排序表,并由排序表提出了新的初始化策略。其次,提出一个新颖的并且有条件调用的动态搜索算法。最后在位置更新上结合瞬态搜索算法(transient search algorithm,TSO),改进勘探和开发步骤的效率,增加解空间的灵活性和多样性,从而使算法能够快速定位到全局最优位置。为了验证算法的性能,实验选取14个UCI的数据集,并且与樽海鞘群算法(SSA)以及近几年樽海鞘群的改进算法等多种优化算法进行比较,结果表明HS-SSA在特征选择上具有更强的竞争力。