A novel approach to designing an adaptive committee applied to predicting company’s future performance

This article presents an approach to designing an adaptive, data dependent, committee of models applied to prediction of several financial attributes for assessing company’s future performance. Current liabilities/Current assets, Total liabilities/Total assets, Net income/Total assets, and Operating Income/Total liabilities are the attributes used in this paper. A self-organizing map (SOM) used for data mapping and analysis enables building committees, which are specific (committee size and aggregation weights) for each SOM node. The number of basic models aggregated into a committee and the aggregation weights depend on accuracy of basic models and their ability to generalize in the vicinity of the SOM node. A random forest is used a basic model in this study. The developed technique was tested on data concerning companies from ten sectors of the healthcare industry of the United States and compared with results obtained from averaging and weighted averaging committees. The proposed adaptivity of a committee size and aggregation weights led to a statistically significant increase in prediction accuracy if compared to other types of committees.     

On robustness of radial basis function network with input perturbation

In this article, we have proposed a methodology for making a radial basis function network (RBFN) robust with respect to additive and multiplicative input noises. This is achieved by properly selecting the centers and widths for the radial basis function (RBF) units of the hidden layer. For this purpose, firstly, a set of self-organizing map (SOM) networks are trained for center selection. For training a SOM network, random Gaussian noise is injected in the samples of each class of the data set. The number of SOM networks is same as the number of classes present in the data set, and each of the SOM networks is trained separately by the samples belonging to a particular class. The weight vector associated with a unit in the output layer of a particular SOM network corresponding to a class is used as the center of a RBF unit for that class. To determine the widths of the RBF units, p-nearest neighbor algorithm is used class-wise. Proper selection of centers and widths makes the RBFN robust with respect to input perturbation and outliers present in the data set. The weights between the hidden and output layers of RBFN are obtained by pseudo inverse method. To test the robustness of the proposed method in additive and multiplicative noise scenarios, ten standard data sets have been used for classification. Proposed method has been compared with three existing methods, where the centers have been generated in three ways: randomly, using k-means algorithm, and based on SOM network. Simulation results show the superiority of the proposed method compared to those methods. Wilcoxon signed-rank test also shows that the proposed method is statistically better than those methods.

     

A detection method for intronic snoRNA genes using extended-weight-updating SOM with appearance probability of bases

Small nucleolar RNAs (snoRNAs) are known that they will participate with RNA modification. However, detail functions of snoRNAs have not been clear still yet. In order to make clear functions of snoRNA, finding more snoRNAs and studying their works in cells are required. In this paper, we propose a method to detect snoRNA genes using extended-weight-updating self-organizing map (eSOM). An input vector to eSOM consists of a feature vector and a target vector. Different from a conventional SOM, a winner node for an input vector is decided by the feature vector only, and all the weights around the winner node are updated to be close to the input vector. We employ bases appearance probabilities and complementary base pair ratio for a feature vector. A target vector is a flag which is 1.0 or 0.0 for a positive or a negative sample, respectively. Experimental results showed our method achieved 91 and 93 % detection ratio for boxC/D and boxH/ACA type snoRNA genes, respectively.     

自组织映射网络的可视化研究

提出了一种自组织映射网络训练结果的可视化方法—距离映射法,该方法根据输入向量与竞争层神经元权向量距离的大小来计算相似度,然后对所有相似度与对应神经元坐标之积求和,把输入向量映射到二维平面。对故障数据的试验分析表明,该方法能提供更加清晰的可视化表示。     

基于SOM网络的股票聚类分析方法

无监督的自组织映射(SOM)神经网络是用于聚类的主要人工神经网络模型之一.在SOM网络的基础上改进了网络中的邻域函数,并将其用于对股票进行分析和选择,得到了令人满意的结果.为了提高解的精度,避免多个输入样本映射到同一输出节点还提出了禁忌映射的方法.数值模拟表明该模型对于上市公司的聚类结果令人满意,对于股民客观、准确地选出真正具有投资价值的股票具有指导意义.     

Online discovery of Heart Rate Variability patterns in mobile healthcare services

Recent years, advances in day-to-day wearable sensors have led to the development of low powered physiological sensor platforms, which can be integrated in body area networks, a new enabling technology for real-time health monitoring. The bottleneck in health state awareness is the algorithm that has to interpret the sensor data. Nowadays Coronary Heart Disease (CHD) is still the leading cause of death. Many classification techniques such as decision tree and neural networks proposed for an early detection of individual at risk for CHD are not able to continuously detect heart state based on sensor data stream. In this study, we propose an online three-layer neural network to recognize Heart Rate Variability (HRV) patterns related to CHD risk in consideration of daily activities. ECG sensor data is preprocessed using Poincaré plot encoding. Incremental learning is utilized to train the network with new data without forgetting the previously learned patterns. The algorithm is named Poincaré-based HRV patterns discovering Incremental Artificial neural Network (PHIAN). When a sample is presented, the nodes in the hidden layer of PHIAN compete for determining the node with the highest similarity to the input. Error variables associated with the neuron units are used as criteria for new node insertion in hopes of allowing the network to learn new patterns and reducing classification error. However, the node insertion has to be stopped in the overlapping decision areas. We suppose that the overlaps between classes have lower probability than the centric part of the classes. Therefore, after a period of learning we remove the nodes with no neighbor. Plus, the error probability density is taken into account instead of input probability density. Finally, the predictive capability of PHIAN is compared with three previous classification models, namely Self-Organizing Map (SOM), Growing Neural Gas (GNG), and Multilayer Perceptron (MLP) in terms of classification error and network structure. The results show that PHIAN outperforms the existing techniques. Our proposed model can be efficiently applied to early detection of abnormal conditions and prevent the abnormal becoming serious.     

基于改进的GHSOM网络预测客户欺诈行为

生长、分级的自组织映射(Growing Hierarchical Self-Organizing Map,GHSOM)网络是自组织映射(Self-Organizing Map,SOM)网络的一种变体,它不仅具备了SOM网络可解释性强的优点,同时采用多层分级的结构,不需要预先定义好网络的结构和尺寸,解决了SOM由于竞争层神经元过多造成的训练时间过长的问题,却忽略了对样本向量各个分量在模型中重要性的分析,因此将一种新的输入模式分量和映射单元权向量之间的灰关联度引入到网络权值的调整过程中,对GHSOM算法进行了改进。运用于对电信客户行为的分类,从中获取了预测欺诈客户的关键指标,大大降低了输入样本的维度。结果显示,采用改进后的GHSOM算法降维后,分类正确率仍然可以达到94.59%。     

Vector quantizer of medical image using wavelet transform and enhanced SOM algorithm

Vector quantizer takes care of special image features like edges, and it belongs to the class of quantizers known as the second-generation coders. This paper proposes a novel vector quantization method using the wavelet transform and the enhanced SOM algorithm for the medical image compression. We propose the enhanced self-organizing algorithm to resolve the defects of the conventional SOM algorithm. The enhanced SOM, at first, reflects the error between the winner node and the input vector to the weight adaptation by using the frequency of the selection of the winner node. Secondly, it adjusts the weight in proportion to the present weight change and the previous one as well. To reduce the blocking effect and the computation requirement, we construct training image vectors involving image features by using the wavelet transform and apply the enhanced SOM algorithm to them for generating a well-defined codebook. Our experimental results have shown that the proposed method energizes the compression ratio and the decompression quality.     

TCSOM: Clustering Transactions Using Self-Organizing Map

Self-Organizing Map (SOM) networks have been successfully applied as a clustering method to numeric datasets. However, it is not feasible to directly apply SOM for clustering transactional data. This paper proposes the Transactions Clustering using SOM (TCSOM) algorithm for clustering binary transactional data. In the TCSOM algorithm, a normalized Dot Product norm based dissimilarity measure is utilized for measuring the distance between input vector and output neuron. And a modified weight adaptation function is employed for adjusting weights of the winner and its neighbors. More importantly, TCSOM is a one-pass algorithm, which is extremely suitable for data mining applications. Experimental results on real datasets show that TCSOM algorithm is superior to those state-of-the-art transactional data clustering algorithms with respect to clustering accuracy.     

多获胜节点SOM及其在股票分析中的应用

为了增强自组织映射(self-organizing map,SOM)网络的动态竞争和聚类能力,提高解的精度,在无监督的SOM神经网络的基础上,通过拓广获胜节点的数量,改进网络中的邻域函数和连接权函数等方法,提出具有多获胜节点的SOM模型.为了避免多个输入样本映射到同一个输出节点,还提出了禁忌映射的方法.为了验证所提出的方法的有效性,以股票的聚类分析为实例,对该方法进行了检验.通过对每股收益、每股净资产、净资产收益率、每股经营性现金流量及净利润等5项反映上市公司综合盈利能力的财务指标进行了模拟实验,所得的数值结果表明,在标准SOM及所提出的几种多获胜节点SOM网络模型中,具有双获胜节点(SOM with 2 winners,SOM2W)的网络模型获得了最好的聚类效果.结合实验结果对网络模型的进一步分析也表明,SOM2W的聚类能力优于标准SOM及其他网络模型.该模型为股票的分析和选择提供了一种可行的途径,在金融领域具有潜在的应用价值.     

Fuzzy integral-based perceptron for two-class pattern classification problems

The single-layer perceptron with single output node is a well-known neural network for two-class classification problems. Furthermore, the sigmoid or logistic function is usually used as the activation function in the output neuron. A critical step is to compute the sum of the products of the connection weights with the corresponding inputs, which indicates the assumption of additivity among individual variables. Unfortunately, because the input variables are not always independent of each other, an assumption of additivity may not be reasonable enough. In this paper, the inner product can be replaced with an aggregation value obtained by a useful fuzzy integral by viewing each of the connection weights as a value of a λ-fuzzy measure for the corresponding variable. A genetic algorithm is then employed to obtain connection weights by maximizing the number of correctly classified training patterns and minimizing the errors between the actual and desired outputs of individual training patterns. The experimental results further demonstrate that the proposed method outperforms the traditional single-layer perceptron and performs well in comparison with other fuzzy or non-fuzzy classification methods.     

SOM神经网络在独立学院招生决策中的应用

自组织映射(SOM)是一种竞争型无指导学习的神经网络方法。SOM神经网络已广泛地应用于模式聚类、模式识别、拓扑不变性映射等方面。本文利用SOM对中国31个省份进行聚类分析,建立独立学院招生决策模型。首先,选取各省份的报到率、第一志愿率和人均GDP等作为SOM神经网络的输入模式;然后,用SOM进行聚类;最后,对聚类结果进行分析得出各类的生源地特征和等级。实验结果表明,利用SOM对生源地进行聚类分析是可行的、有效的,可以避开人的主观因素,更迅速客观地得到聚类结果。它为独立学院编制招生计划和招生宣传方案提供了一种新的参考依据,在独立学院招生领域具有较好的应用前景。     

Different Aspects of Clustering The Self-Organizing Maps

Self-organizing map (SOM) is an artificial neural network tool that is trained using unsupervised learning to produce a low dimensional representation of the input space, called a map. This map is generally the object of a clustering analysis step which aims to partition the referents vectors (map neurons) into compact and well-separated groups. In this paper, we consider the problem of the clustering SOM using different aspects: partitioning, hierarchical and graph coloring based techniques. Unlike the traditional clustering SOM techniques, which use k-means or hierarchical clustering, the graph-based approaches have the advantage of providing a partitioning of the SOM by simultaneously using dissimilarities and neighborhood relations provided by the map. We present the experimental results of several comparisons between these different ways of clustering.     

Piecewise Linear Projection Based on Self-Organizing Map

A piecewise linear projection algorithm, based on kohonen's Self-Organizing Map, is presented. Using this new algorithm, neural network is able to adapt its neural weights to accommodate with input space, while obtaining reduced 2-dimensional subspaces at each neural node. After completion of learning process, first project input data into their corresponding 2-D subspaces, then project all data in the 2-D subspaces into a reference 2-D subspace defined by a reference neural node. By piecewise linear projection, we can more easily deal with large data sets than other projection algorithms like Sammon's nonlinear mapping (NLM). There is no need to re-compute all the input data to interpolate new input data to the 2-D output space.     

有监督SOM神经网络在入侵检测中的应用

为提高自组织特征映射(SOM)神经网络的分类性能,提出一种有监督SOM神经网络(SSOM)。在输入层和竞争层的基础上增加输出层,根据输入样本的不同预测类别,选取不同的公式调整权值,并训练网络。通过2个权值的组合,实现对样本类别的回归和统计。基于KDD CUP99入侵检测数据集的实验结果表明,与其他SOM网络相比,SSOM具有更好的分类性能和更高的入侵检测率。     

Review of the Self-Organizing Map (SOM) approach in water resources: Commentary

We provide some additional input and perspectives on Kalteh et al's review of the Self-Organizing Map (SOM) approach (Environ. Model. Softw. (2008), 23, 835–845). Map size selection is a key issue in SOM applications. Although there is no theoretical principle to determine the optimum map size, quantitative indicators such as quantization error, topographic error and eigenvalues have proven to be relevant tools to determine the optimal number of map units. Second, one of the most innovative applications of the SOM is the possibility of introducing a set of variables (e.g., biological) into a SOM previously trained with other variables (e.g. environmental). This can be achieved by calculating the mean value of each environmental variable in each output neuron of a SOM trained with biological variables, or by using a mask function to give a null weight to the biological variables, whereas environmental variables are given a weight of 1 so that the values for biological variables are visualized on a SOM previously trained with environmental variables only. We conclude that our different levels of expertise represent an opportunity for stimulating cross-fertilisation in the vast field of water research rather than simply yielding a collection of case studies to be re-examined.     

自组织特征映射神经网络的改进及应用研究

为了提高自组织特征映射(SOM)神经网络学习速度及分类精度,对初始连接权值及竞争层神经元数的确定方法进行改进。提出用聚类方法确定初始权值的新方法,还提出了采用聚类数与邻域之和确定竞争层神经元数的方法,并给出了改进后的SOM分类算法。将改进的SOM网络用于储粮害虫分类,采用留一方法进行分类验证实验。仿真结果表明,改进后的SOM网络在学习速度和分类精度方面都有明显提高,证明了该方法的有效性。     

Analysing coastal ocean model outputs using competitive-learning pattern recognition techniques

To assist in interpreting the hydrodynamics of a complex coastal environment, a Self Organizing Map (SOM) has been constructed using output from a three-dimensional hydrodynamic model of the Huon-D'Entrecasteaux region in South-East Tasmania, over a one-year period. Interpretation of the SOM enabled nine characteristic or prototype states to be identified. As expected, the dominant forcing mechanisms were freshwater input via riverine discharge and input from oceanic waters. While these mechanisms are well understood, subtle features associated with the interaction of the two forcing mechanisms and the transitions between meta-stable states, were revealed by visualizing the SOM output. Further investigation was undertaken to determine how effective the SOM would be in identifying these prototype states given sensor data from a sensor network being designed for future deployment within the region. This research has demonstrated that SOM analysis can be a useful tool for identifying and interpreting patterns in large oceanographic datasets.     

The effect of sample size on the extended self-organizing map network—A market segmentation application

Kohonen's self-organizing map (SOM) network maps input data to a lower dimensional output map. The extended SOM network further groups the nodes on the output map into a user specified number of clusters. Kiang, Hu and Fisher used the extended SOM network for market segmentation and showed that the extended SOM provides better results than the statistical approach that reduces the dimensionality of the problem via factor analysis and then forms segments with cluster analysis. In this study, we examined the effect of sample size on the extended SOM compared to that on the factor/cluster approach. Two sampling schemes, one with random sampling and the other one with proportionate sampling were used. Comparisons were made using the correct classification rates between the two approaches at various sample sizes. Unlike statistical models, neural networks are not dependent on statistical assumptions. Thus, the results for neural network models are stable across sample sizes but sensitive to initial weights and model specifications.     

多个水下机器人动态任务分配和路径规划的信度自组织算法

针对多个水下机器人(autonomous underwater vehicles,AUVs)动态任务分配和路径规划速度跳变问题,引入栅格信度函数概念,给出一种改进的栅格信度自组织(belief function self-organizing map,BFSOM)算法.目的是控制一组AUV有效地到达所有指定的目标位置,同时保证AUV能够自动的避开障碍物.首先,自组织神经网络(self-organizing map,SOM)算法对多AUV系统进行任务分配,使得每个目标位置都有一个AUV去访问.整个分配过程包括定义SOM神经网络的初始权值、获胜者选择、邻域函数的计算3个步骤;其次,根据栅格信度函数和环境信息更新SOM获胜神经元的权值,使得每个AUV在访问对应目标的过程中能够自动避障并且克服速度跳变,实现AUV自动有效路径规划.最后,通过仿真实验证明了本文提及算法的有效性.