Expanding Self-Organizing Map for data visualization and cluster analysis

Self organized mapping of data clusters to neuron groups

In this paper, we cluster and analyze data from the past embedded software development projects using self-organizing maps (SOMs)[9] that are a type of artificial neural networks that rely on unsupervised learning. The purpose of the clustering and analysis is to improve the accuracy of predicting the number of errors. A SOMproduces a low-dimensional, discretized representation of the input space of training samples; these representations are called maps. SOMs are useful for visualizing low-dimensional views of high-dimensional data, a multidimensional scaling technique. The advantages of SOMs for statistical applications are as follows: (1) data visualization, (2) information processing on association and recollection, (3) summarizing large-scale data, and (4) creating nonlinear models. To verify our approach, we perform an evaluation experiment that compares SOM classification to product type classification using Welch’s t-test for Akaike’s Information Criterion (AIC). The results indicate that the SOM classification method is more contributive than product type classification in creating estimation models, because the mean AIC of SOM classification is statistically significantly lower.KeywordsSelf-organizing mapsclusteringembedded software development

Artificial neural networks assessing adolescent idiopathic scoliosis: comparison with Lenke classification

In this paper, two methods for generating the daily load profile and forecasting in isolated small communities are proposed. In these communities, the energy supply is difficult to predict because it is not always available, is limited according to some schedules and is highly dependent on the consumption behavior of each community member. The first method is proposed to be used before the implementation of the microgrid in the design state, and it includes a household classifier based on a Self Organizing Map (SOM) that provides load patterns by the use of the socio-economic characteristics of the community obtained in a survey. The second method is used after the implementation of the microgrid, in the operation state, and consists of a neural network with on-line learning for the load forecasting. The neural network model is trained with real-data of load and it is designed to stay adapted according to the availability of measured data. Both proposals are tested in a real-life microgrid located in Huatacondo, in northern Chile (project ESUSCON). The results show that the estimated daily load profile of the community can be very well approximated with the SOM classifier. On the other hand, the neural network can forecast the load of the community reasonably well two-days ahead. Both proposals are currently being used in a key module of the energy management system (EMS) in the real microgrid to optimize the real uninterrupted load for 24-hour energy supply service.

It has been hypothesized that the different appearance of ocular dominance bands in the cat and the monkey is a consequence of the different mapping geometries in these species (LeVay et al. 1985; Anderson et al. 1988). Here I investigate the impact of areal geometries on the preferred direction of ocular dominance bands in two adaptive map formation models, the self-organizing feature map and the elastic net algorithm. In the case of the self-organizing feature map, the occurrence of instabilities that correspond to ocular dominance bands can be analytically investigated. The instabilities automatically yield stripes of correct orientation. These analytic results are complemented by simulations. In the case of the elastic net algorithm, simulations reveal two different parameter regimes of the algorithm, only one of which leads to stripes of correct orientation. The results suggest that neighborhood preservation in visual maps is enforced in the backward direction, such that neighboring cells in the cortex have neighboring receptive fields, and not vice versa.

We have recently suggested using clustering of variables (CLoVA) based on unsupervised pattern recognition for partitioning variables into informative and redundant ones. Because data clustering plays a central role in CLoVA, in the present study, we compared the efficiency of different clustering methods including the Kohonen self‐organizing map (SOM), principal component analysis, fuzzy c‐means clustering, K‐means clustering, and hierarchical cluster analysis for clustering spectroscopic data and molecular descriptors to build multivariate calibration and quantitative structure–activity (QSAR) models. To investigate which clustering methods are more efficient for CLoVA, four data sets (three spectroscopic and one QSAR) were analyzed. Most of the CLoVA‐based models obtained by SOM resulted in the least root mean square errors of cross‐validation and prediction, suggesting a higher efficiency of SOM for clustering variables. In all cases, the results obtained by the CLoVA‐based method were compared with those obtained by conventional principal component regression as well as genetic algorithm and successive projection algorithm partial least square regression. Interestingly, models produced by the CLoVA‐based method were more predictive with respect to that of the other methods, as indicated by the lowest root mean square error of prediction. Copyright © 2013 John Wiley & Sons, Ltd.

OSCAR is a web platform for cluster and cross-species analysis of microarray data. It provides a comprehensive but friendly environment to both users and algorithm developers. For users, OSCAR provides cluster tools for both single and multiple species data, together with interactive analysis features. For single species data, OSCAR currently provides Hierarchical Clustering, K-means, partition around medoids (PAM), Self-Organizing Map (SOM), Tight Clustering and a novel algorithm called ‘Consensus Tight-clustering’. The new Consensus Tight-clustering algorithm delivers robust gene clusters and its result is more resistant to false positives than other state-of-the-art algorithms. For cross-species data analysis, OSCAR provides two novel computational tools: ‘coherentCluster’, ‘coherentSubset’ and a novel visualization tool: ‘comparative heatmap’. Applying the coherentCluster algorithm to human and fly aging data, we identified several coherent clusters of genes, which share co-regulation patterns that are highly correlated with the aging process in both of the two species. One coherent cluster suggests insulin receptor (INSR) may regulate Pax6 in both species and across different tissues. Further analysis with human brain expression and pathological data suggests an INSR->Pax6->quinone oxidoreductase (NQO1)->detoxification neuro-protective pathway might be present in aging or diseased brain. For algorithm developers, OSCAR is a plug-and-play platform. With little effort, developers can plug their own algorithms into the OSCAR server without revealing the source codes, which will equip their command line executables with user-friendly interface and interactive analysis capability. In summary, OSCAR initiates an open platform for development and application of clustering and cross-species analysis programs. OSCAR stands for an open system for cluster analysis of microarray data. It is available at: http://biocomp.bioen.uiuc.edu/oscar

Artificial neural network analysis of noisy visual field data in glaucoma

The self-organizing map (SOM) is naturally unsupervised learning, but if a class label is known, it can be used as the classifier. In a SOM classifier, each neuron is assigned a class label based on the maximum class frequency and classified by a nearest neighbor strategy. The drawback when using this strategy is that each pattern is treated by equal importance in counting class frequency regardless of its typicalness. For this reason, the fuzzy class membership can be used instead of crisp class frequency and this fuzzy membership-label neuron provides another perspective of a feature map. This fuzzy class membership can be also used to select training samples in a support vector machine (SVM) classifier. This method allows us to reduce the training set as well as support vectors without significant loss of classification performance.

The comprehension of the behavior of objects in the environment of an automated vehicle is essential information to be extracted out of traffic measurement data. It can be made accessible by applying a data description process consisting of a cluster analysis and subsequent classification step. In this paper, two different algorithms, TRACLUS and Self-Organizing Maps are utilized for a clustering of laser scanner measurement data, recorded by experimental vehicles in real traffic. The results are evaluated and discussed. A remaining open issue of a cluster analysis is the necessity for a classification of clusters, which is missing or solved through manual solutions in related work. We address the matter with a novel approach for an automated classification of clusters using reference trajectories, which is successfully validated for a first exemplary maneuver on highway measurement data.

Spiking Self-organizing Maps for Classification Problem

Concrete pavement joint evaluation involves a number of assessment criteria, such as deflection near the joint, load transfer efficiency of the dowels and severity of voids under the slab. Although there are well defined thresholds for each of these parameters, often there arises a situation where each of the considered parameters lends contradictory assessment that leads to a considerable subjectivity in the evaluation process. A Self-Organizing Map (SOM), an unsupervised learning procedure in artificial neural network, is utilised for the first time to map the joint condition of concrete pavements from Falling Weight Deflectometer (FWD) deflection bowls. A novel methodology is proposed for labelling the network, whereby pavement engineering expertise can be directly used in a SOM for consistent deflection data classification in joint evaluation. The effectiveness of the trained network is demonstrated by using joint assessment parameters; namely, load transfer efficiency (LTE), void intercepts and absolute deflection. The joints were classified as good, marginal or poor. For the three parameters based SOM classification, an accuracy of 65-70% was obtained; this improves to 87.5% when the SOM was trained with 2-parameters (LTE and absolute deflection). However, when the SOM was tested with the data classified as 'good', accuracy improves to around 90%. Therefore, a SOM can be a powerful supplementary tool for a consistent and non-subjective evaluation of concrete pavement joints.

Cluster analysis is an important data mining technique for analyzing large amounts of data, reducing many objects to a limited number of clusters. Cluster visualization techniques aim at supporting the user in better understanding the characteristics and relationships among the found clusters. While promising approaches to visual cluster analysis already exist, these usually fall short of incorporating the quality of the obtained clustering results. However, due to the nature of the clustering process, quality plays an important aspect, as for most practical data sets, typically many different clusterings are possible. Being aware of clustering quality is important to judge the expressiveness of a given cluster visualization, or to adjust the clustering process with refined parameters, among others. In this work, we present an encompassing suite of visual tools for quality assessment of an important visual cluster algorithm, namely, the Self-Organizing Map (SOM) technique. We define, measure, and visualize the notion of SOM cluster quality along a hierarchy of cluster abstractions. The quality abstractions range from simple scalar-valued quality scores up to the structural comparison of a given SOM clustering with output of additional supportive clustering methods. The suite of methods allows the user to assess the SOM quality on the appropriate abstraction level, and arrive at improved clustering results. We implement our tools in an integrated system, apply it on experimental data sets, and show its applicability.

The clustering of high-dimensional data presents a critical computational problem. Therefore, it is convenient to arrange the cluster centres on a grid with a small dimensional space that reduces computational cost and can be easily visualized. Moreover, in real applications there is no sharp boundary between classes, real datasets are naturally defined in a fuzzy context. Therefore, fuzzy clustering fits better for complex real datasets to determine the best distribution. Self-organizing map (SOM) technique is appropriate for clustering and vector quantization of high-dimensional data. In this paper we present a new fuzzy learning approach called FMIG (fuzzy multilevel interior growing self-organizing maps). The proposed algorithm is a fuzzy version of MIGSOM (multilevel interior growing self-organizing maps). The main contribution of FMIG is to define a fuzzy process of data mapping and to take into account the fuzzy aspect of high-dimensional real datasets. This new algorithm is able to auto-organize the map accordingly to the fuzzy training property of the nodes. In the second step, the introduced scheme for FMIG is clustered by means of fuzzy C-means (FCM) to discover the interior class distribution of the learned data. The validation of FCM partitions is directed by applying six validity indexes. Superiority of the new method is demonstrated by comparing it with crisp MIGSOM, GSOM (growing SOM) and FKCN (fuzzy Kohonen clustering network) techniques. Thus, our new method shows improvement in term of quantization error, topology preservation and clustering ability.

This paper presents a cluster analysis method which automatically finds the number of clusters as well as the partitioning of a data set without any type of interaction with the user. The data clustering is made using the self-organizing (or Kohonen) map (SOM). Different partitions of the trained SOM are obtained from different segmentations of the U-matrix (a neuron-distance image) that are generated by means of mathematical morphology techniques. The different partitions of the trained SOM produce different partitions for the data set which are evaluated by cluster validity indexes. To reduce the computational cost of the cluster analysis process this work also proposes the simplification of cluster validity indexes using the statistical properties of the SOM. The proposed methodology is applied in the cluster analysis of remotely sensed images.