Unsupervised Artificial Neural Network Research Articles

GUASOM: Gaia Utility for Analysis and Knowledge Discovery based on Self Organizing Maps

We present a method for knowledge analysis in large astronomical spectrophotometric archives. The method is based on a type of unsupervised learning Artificial Neural Networks named Self-organizing maps (SOMs). SOMs are used to organize the information in clusters of objects, as homogeneously as possible according to their spectral energy distributions, and to project them onto a 2D grid where the data structure can be visualized. Our algorithm has been tested by means of simulated Gaia spectrophotometry (150,000 objects), which is based on SDSS observations and theoretical spectral libraries covering a wide sample of astronomical objects. We demonstrate the usefulness of the method by analyzing over 10,000 objects, mostly fainted objects and unsuccessful observations, that were rejected by the SDSS spectroscopic classification pipeline and thus classified as “UNKNOWN”. This dataset was transformed to Gaia BP and RP format by the use of GOG simulator. GUASOM provides a useful toolbox to study the data distribution in extense archives. Even more, the discovered neighbourhood relationships help to unveil the physical nature of objects never observed before. To this effect, we used the SIMBAD catalog to perform crossmatching with the SDSS astrometry, seeking for more identifications.

Visual topical analysis of Chinese and American Library and Information Science research institutions

Abstract Research institutions play an important role in scientific research and technical innovation. The topical analysis of research institutions in different countries can facilitate mutual learning and promote potential collaboration. In this study, we illustrate how an unsupervised artificial neural network technique Self-Organizing Map (SOM) can be used to visually analyze the research fields of research institutions. A novel SOM display named Compound Component Plane (CCP) was presented and applied to determine the institutions which made significant contributions to the salient research fields. Eighty-seven Chinese and American LIS institutions and the technical LIS fields were taken as examples. Potential international and domestic collaborators were identified based upon their research similarities. An approach of dividing research institutions into clusters was proposed based on their geometric distances in the SOM display, the U -matrix values and the most salient research topics they involved. The concepts of swarm institutions, pivots and landmarks were also defined and their instances were identified.

Current Scope, Case Studies and Future Directions of Ecological Informatics

Ecological data management, analysis and synthesis as well as information processing and transfer in ecology are considered feature areas of ecological informatics. The paper presents case studies for analysis and synthesis of ecological data by means of unsupervised artificial neural networks and hybrid evolutionary algorithms to assist in computational bioindication of wetland water quality and early warning of cyanobacteria blooms in lakes and reservoirs. Integrated analysis and synthesis of ecological and genomics data, hybrid model libraries generic for ecosystem categories as well as internet-based data and model sharing are identified as key research areas of ecological informatics.

Application of Self-Organizing Artificial Neural Networks on Simulated Diffusion Tensor Images

Diffusion tensor magnetic resonance imaging (DTMRI) as a noninvasive modality providing in vivo anatomical information allows determination of fiber connections which leads to brain mapping. The success of DTMRI is very much algorithm dependent, and its verification is of great importance due to limited availability of a gold standard in the literature. In this study, unsupervised artificial neural network class, namely, self-organizing maps, is employed to discover the underlying fiber tracts. A common artificial diffusion tensor resource, named “phantom images for simulating tractography errors” (PISTE), is used for the accuracy verification and acceptability of the proposed approach. Four different tract geometries with varying SNRs and fractional anisotropy are investigated. The proposed method, SOFMAT, is able to define the predetermined fiber paths successfully with a standard deviation of (0.8–1.9) × 10−3depending on the trajectory and the SNR value selected. The results illustrate the capability of SOFMAT to reconstruct complex fiber tract configurations. The ability of SOFMAT to detect fiber paths in low anisotropy regions, which physiologically may correspond to either grey matter or pathology (abnormality) and uncertainty areas in real data, is an advantage of the method for future studies.

A Neural Network Model for Classifying Olive Farms

The application of web marketing and definition of corporate strategies has become common practice in all companies, together with the use of mathematical models as a tool for planning and studying the dynamics of communication within the market. We apply an unsupervised artificial neural network for the classification of a series of olive farms to try to determine which features are most rewarding from the point of view of the communication strategies and market (including the identification of new situations and decision making). The objective is to identify and group companies that have similar characteristics through a set of common indicators and create a rating for defining which companies are the best performing and how companies in the sector are related. This work is made possible by the use of a computer software designed specifically for the olive oil sector, which examines many aspects of business life and also implements the platform among businesses, each other and the market.

Application of artificial neural network (ANN)–self-organizing map (SOM) for the categorization of water, soil and sediment quality in petrochemical regions

► The physico-chemical properties and heavy metals in environmental materials from the Niger Delta area were assessed. ► The SOM was used as a powerful visualization tool to identify trends in the dataset. ► Preliminary diagnoses of the quality of locational environmental materials are effectively carried out using the SOM algorithm to develop the c-planes . ► Areas with high concentrations of pollutants were easily identified from the c-planes . ► These stations from the Niger Delta should be classified as high priority sites. The utilization of mathematical and computational tools for pollutant assessment frameworks has become increasingly valuable due to the capability to interpret integrated variable measurements. Artificial neural networks (ANNs) are considered as dependable and inexpensive techniques for data interpretation and prediction. The self-organizing map (SOM) is an unsupervised ANN used for data training to classify and effectively recognize patterns embedded in the input data space. Application of SOM–ANN is useful for recognizing spatial patterns in contaminated zones by integrating chemical, physical, ecotoxicological and toxicokinetic variables in the identification of pollution sources and similarities in the quality of the samples. Water ( n = 11), soil ( n = 38) and sediment ( n = 54) samples from four areas in the Niger Delta (Nigeria) were classified based on their chemical, toxicological and physical variables applying the SOM. The results obtained in this study provided valuable assessment using the SOM visualization capabilities and highlighted zones of priority that might require additional investigations and also provide productive pathway for effective decision making and remedial actions.

Performance enhanced PSO-based modified Kohonen neural network for retinal image classification

Image classification is one of the significant applications in the field of ophthalmology for abnormality detection in retinal images. Image classification is a pattern recognition technique in which abnormal retinal images are categorized into different groups based on similarity measures. Accuracy and convergence rate are the important parameters of this automated diagnostic system. Artificial neural networks (ANNs) are widely used for automated image analysis systems. Kohonen neural networks (KNNs) are one of the prime unsupervised ANNs suitable for image processing applications. Besides the numerous advantages, KNNs suffer from two drawbacks: (a) lack of standard convergence conditions and (b) less accurate results. In this study, a novel approach is adopted to eliminate these disadvantages by performing suitable modifications in the conventional KNN. Initially, the fuzzy approach is an integrated one within KNN in the training algorithm to overcome the convergence difficulties. Second, a particle swarm optimization algorithm is used in feature selection for better accuracy. This proposed approach is tested on four different abnormal retinal image categories. The system is analyzed using several performance measures and the experimental results suggest promising results for the proposed system. Comparative analyses with other systems are also presented to show the superior nature of the proposed system.

Ecological exergy as an indicator of land-use impacts on functional guilds in river ecosystems

The cumulative effect of land-use changes is one of the most important factors contributing to the continuous deterioration of river ecosystems. We used ecological exergy to evaluate the impacts of land-use changes on functional guilds of benthic macroinvertebrates. We classified 353 sampling sites into 3 groups based on land-use types: forested, agricultural, and urban rivers. For each sampling site, we calculated ecological exergy based on 5 trophic groups of macroinvertebrates. Differences in exergy, specific exergy, and structural metrics (i.e. species richness and Shannon index) suggested that land-use type was an important determinant of the composition of macroinvertebrate communities. Exergy values of the functional feeding groups and trophic groups were used as input data to train self-organizing maps – unsupervised artificial neural networks. The results showed that functional guilds responded differently to different land-use types: scrapers and carnivores dominated the forested rivers, whilst predators and omnivores, and gatherer-collectors and detritivores dominated agricultural and urban rivers, respectively. These results suggest that ecological exergy can be used as a functional bioassessment indicator to evaluate river condition. A generalized additive model and random forest also highlighted that a combination of both conventional structural indicators (e.g. species richness and Shannon index) and novel functional indicators (e.g. exergy and specific exergy) can be used to assess biotic integrity.

Hybrid modeling to predict the economic feasibility of mining undiscovered porphyry copper deposits

Few studies attempt to model the economic feasibility of mining undiscovered mineral resources given the sparseness of data; and the coupled, nonlinear, spatial, and temporal relationships among variables. In this study, a type of unsupervised artificial neural network, called a self-organized map (SOM), is trained using data from 203 porphyry copper deposit sites across the world. The sparse data set includes one dependent variable indicating the economic feasibility, and seventy two independent variables from categories describing characteristics of mining method, metallurgy, dimensions, economics, and amount. Analysis of component planes reveals relations and strengths in the underlying SOM multivariate density function which are used to impute missing values. Application of the Davies–Bouldin criteria to k-means clusters of SOM neurons identified 14 regional economic resource units (conceptual models). A best subsets approach applied to median values from these models identified 20 statistically significant combinations of variables. During model fitting by the multiple linear regression technique, only four of the empirical models had variables that were all significant at the 95% confidence level. The best model explained 98% of the variability in economic feasibility and incorporated variables describing distance to natural gas, road, and water; and the total amount of resources. This model was independently validated by comparing predictions of economic feasibility at 68 mine sites not included in the training data. Eighty-four percent of the reported economic feasibility is correctly predicted with 8 false positives and 2 false negative. We demonstrate the application of this model to a permissive copper porphyry tract that crosses a portion of British Columbia and Yukon territories of Canada. The proposed hybrid approach provides an alternative modeling paradigm for translating estimates of contained metal into meaningful societal measures.

Characterizing differential responses of benthic macroinvertebrate communities to floods and droughts in three different stream types using a Self‐Organizing Map

ABSTRACTWe investigated the spatial and temporal changes of benthic macroinvertebrate communities in response to floods and droughts in three different stream types: perennial, intermittent, and ephemeral. Benthic macroinvertebrates were collected monthly (or biweekly) at six different sampling sites from July 2006 to January 2008. Species richness and abundance were extremely low at all sampling sites during the flooding period, but the communities subsequently recovered rapidly within 1 month. The effects of floods and droughts on benthic macroinvertebrates manifested differently among the three different stream types. Species richness and abundance had a negative relationship with the amount of precipitation in the perennial stream type, whereas no significant relationship was observed in the ephemeral stream type. A Self‐Organizing Map, an unsupervised artificial neural network, classified the samples into four clusters based on the similarities of community composition, reflecting the different effects of floods and droughts on the different stream types. Our results suggest that floods and droughts are the defining characteristics of streams, spatial heterogeneity (i.e. the perennial stream type) can decrease the effect of temporal heterogeneity, and both spatial and temporal heterogeneities are important for structuring the environmental condition and benthic macroinvertebrate assemblages. Copyright © 2012 John Wiley & Sons, Ltd.

Maximal Ratio Combining using Self Organizing Map in Wireless Channels

The work is related to the use of Self Organizing Map (SOM) which is a type of unsupervised Artificial Neural Network (ANN), as an aid to Maximal Ratio Combining (MRC) in order to improve bit error rate (BER) values of demodulated signals in wireless channels that have both Gaussian and multipath fading characteristics. Among the architectures and algorithms suggested for ANN, the SOM has the special property of effectively creating spatially organised “ internal representations” of various features of input signals and their

Classification of chickpea seeds using supervised and unsupervised artificial neural networks

As a result of varietal variability in agricultural crops, identification of seed varieties is an important problem. In this study, the ability of Artificial Neural Networks (ANN) in classification of chickpea seeds varieties was considered based on morphological properties of seeds. Experimentally, the seven morphological feature of 400 seeds (including four varieties; Kaka, Piroz, Ilc and Jam) were obtained. Using a combination of input variables, a database of 400 patterns was obtained for the development of ANN models. For comparing the supervised and unsupervised artificial neural networks in classification, the back propagation algorithm (BP) and self-organizing map (SOM) were used for classification. The results of this study showed that unsupervised artificial neural network has a better performance (with 79% accuracy and R2 = 0.8455) in classification of chickpea varieties rather than supervised artificial neural networks (with 73% accuracy and R2= 0.8236). Key words: Back propagation, classification accuracy, morphological properties, self-organizing map, variety.

Applying artificial neural network theory to exploring diatom abundance at tropical Putrajaya Lake, Malaysia

This article explores the relationship between diatom abundance and water quality variables in tropical Putrajaya Lake based on limnological data collected from 2001 to 2006, using supervised and unsupervised artificial neural networks (ANNs). Recurrent artificial neural network (RANN) was used for the supervised ANNs and Kohonen Self Organizing Feature Maps (SOMs) for the unsupervised ANNs. The RANN was developed for the prediction of diatom abundance using variables selected by sensitivity analysis (water temperature, pH, dissolved oxygen, and turbidity). The RANN model performance was measured using root mean squared error (19.0 cell/mL) and the r-value (0.7). SOM was used in this study for classification and clustering of diatom abundance in relation to selected water quality variables and was validated using a sensitivity curve of diatom abundance over the selected variable range generated from RANN. SOM has been employed in this study for pattern discovery of diatom abundance at Putrajaya Lake. The extracted patterns of diatom abundance in terms of propositional IF…else rules were tested and yielded an accuracy rate of 87%.

Hybrid analysis of multiaxis electromagnetic data for discrimination of munitions and explosives of concern

SUMMARY The remediation of land containing munitions and explosives of concern, otherwise known as unexploded ordnance, is an ongoing problem facing the U.S. Department of Defense and similar agencies worldwide that have used or are transferring training ranges or munitions disposal areas to civilian control. The expense associated with cleanup of land previously used for military training and war provides impetus for research towards enhanced discrimination of buried unexploded ordnance. Towards reducing that expense, a multiaxis electromagnetic induction data collection and software system, called ALLTEM, was designed and tested with support from the U.S. Department of Defense Environmental Security Technology Certification Program. ALLTEM is an on-time time-domain system that uses a continuous triangle-wave excitation to measure the target-step response rather than traditional impulse response. The system cycles through three orthogonal transmitting loops and records a total of 19 different transmitting and receiving loop combinations with a nominal spatial data sampling interval of 20 cm. Recorded data are pre-processed and then used in a hybrid discrimination scheme involving both data-driven and numerical classification techniques. The data-driven classification scheme is accomplished in three steps. First, field observations are used to train a type of unsupervised artificial neural network, a self-organizing map (SOM). Second, the SOM is used to simultaneously estimate target parameters (depth, azimuth, inclination, item type and weight) by iterative minimization of the topographic error vectors. Third, the target classification is accomplished by evaluating histograms of the estimated parameters. The numerical classification scheme is also accomplished in three steps. First, the Biot–Savart law is used to model the primary magnetic fields from the transmitter coils and the secondary magnetic fields generated by currents induced in the target materials in the ground. Second, the target response is modelled by three orthogonal dipoles from prolate, oblate and triaxial ellipsoids with one long axis and two shorter axes. Each target consists of all three dipoles. Third, unknown target parameters are determined by comparing modelled to measured target responses. By comparing the rms error among the self-organizing map and numerical classification results, we achieved greater than 95 per cent detection and correct classification of the munitions and explosives of concern at the direct fire and indirect fire test areas at the UXO Standardized Test Site at the Aberdeen Proving Ground, Maryland in 2010.

Self-organising map rainfall-runoff multivariate modelling for runoff reconstruction in inadequately gauged basins

Water resources assessment activities in inadequately gauged basins are often significantly constrained due to the insufficiency or total lack of hydro-meteorological data, resulting in huge uncertainties and ineffectual performance of water management schemes. In this study, a new methodology of rainfall-runoff modelling using the powerful clustering capability of the self-organising map (SOM), unsupervised artificial neural networks, is proposed as a viable approach for harnessing the multivariate correlation between the typically long record rainfall and short record runoff in such basins. The methodology was applied to the inadequately gauged Osun basin in southwest Nigeria for the sole purpose of extending the available runoff records and, through that, reducing water resources planning uncertainty associated with the use of short runoff data records. The extended runoff records were then analysed to determine possible abstractions from the main river source at different exceedance probabilities. This study demonstrates the successful use of emerging tools to overcome practical problems in sparsely gauged basins.

A new geometric model for clustering high-resolution satellite images

Image segmentation is a central process in image processing. There are many segmentation methods such as region growing, edge detection, split and merge and artificial neural networks (ANNs). However, the most important and popular are clustering methods. Normally, clustering methods select cluster centres randomly to segment an image into disjoint and homogeneous regions. The use of random cluster centres without a priori knowledge leads to degradation in the accuracy of the obtained results. However, combined with edge detection, shape representation can help in improving the clustering methods. The improvement is obtained by knowing the optimal location of the cluster centres at the beginning of the image segmentation process. In this article, a new geometric model for high-resolution satellite image segmentation is implemented that can overcome the problem encountered in random clustering processes. The proposed model uses Canny–Deriche edge detection and the modified non-uniform rational B-spline (NURBS) methods to generate the control points of the edges. These points are used to identify cluster centres that are necessary to create the population of the hybrid dynamic genetic algorithm (HDGA). The new geometric model is compared with the self-organizing maps (SOMs) method, which is an efficient unsupervised ANN method. Two experiments are conducted using high-resolution satellite images, and the results prove the high accuracy and reliability of the new evolutionary geometric model.

Evaluation of Changes in Effluent Quality from Industrial Complexes on the Korean Nationwide Scale Using a Self-Organizing Map

One of the major issues related to the environment in the 21st century is sustainable development. The innovative economic growth policy has supported relatively successful economic development, but poor environmental conservation efforts, have consequently resulted in serious water quality pollution issues. Hence, assessments of water quality and health are fundamental processes towards conserving and restoring aquatic ecosystems. In this study, we characterized spatial and temporal changes in water quality (specifically physico-chemical variables plus priority and non-priority pollutants) of discharges from industrial complexes on a national scale in Korea. The data were provided by the Water Quality Monitoring Program operated by the Ministry of Environment, Korea and were measured from 1989 to 2008 on a monthly basis at 61 effluent monitoring sites located at industrial complexes. Analysis of monthly and annual changes in water quality, using the seasonal Mann-Kendall test, indicated an improvement in water quality, which was inferred from a continuous increase in dissolved oxygen and decrease in other water quality factors. A Self-Organizing Map, which is an unsupervised artificial neural network, also indicated an improvement of effluent water quality, by showing spatial and temporal differences in the effluent water quality as well as in the occurrence of priority pollutants. Finally, our results suggested that continued long-term monitoring is necessary to establish plans and policies for wastewater management and health assessment.

An Intelligent Active Queue Management Technique for congestion control

Congestion an major problem in today’s internet traffic had solution with TCP/IP congestion control mechanism. The active queue management (AQM) schemes stabilized the queue oscillations. Earlier RED AQM technique maintained the queue stability in which parameter setting was difficult. Hence a intelligent technique to stabilize the queue in the rapid growing traffic in internet was required. This paper proposes new unsupervised artificial neural network architecture with competitive learning mechanism. Learning vector quantization (LVQ) stabilizes the queue and reduces the queue oscillation. The results are compared with the Kohonen RED (KRED) and Modified Kohonen RED (MKRED) and prove that the proposed LVQ architecture

Infilling of missing rainfall and streamflow data in the Shire River basin, Malawi – A self organizing map approach

A major requirement for the assessment, development and sustainable use of water resources is the availability of good quality hydrological time series data of sufficiently long duration. However, it is not uncommon to find data that are riddled with gaps, characterized by questionable quality and short durations. Sometimes, the data are just not available. Such situations are most prevalent in developing countries and the consequence is a high degree of uncertainty in the assessed characteristics of water management schemes and ultimately its ineffectual performance. Thus dealing with these problems is an important exercise in hydrological analyses. This paper focuses on the multivariate infilling of gaps for rainfall and streamflow data in the Shire River basin in Malawi, using a self organizing map (SOM) approach, which is a form of unsupervised artificial neural networks. The results show that this approach can produce reliable estimates of hydro-meteorological data thus offering promise for reducing the uncertainties associated with the use of insufficient data for water resources assessment.

Long-term changes and recurrent patterns in fisheries landings from Large Marine Ecosystems (1950–2004)

The regional dynamics of industrial fisheries within Large Marine Ecosystems (LMEs) boundaries were investigated by means of a historical-descriptive approach. Landings data from the Sea Around Us Project database were used to detect trends in total yields and variations in landings composition by functional groups over time. The temporal and spatial scales covered by this study allowed general issues to be addressed such as the detection of recurrent patterns and synchronies in fisheries landings. An unsupervised artificial neural network, Self Organizing Map (SOM), is used as a tool to analyze fisheries landings composition variation over five decades in 51 LMEs all over the world. From the historical analysis of “fishing behaviors” within LMEs a broad distinction between two main types of fisheries emerged: (1) small and medium pelagics fisheries, with stable compositions or cyclic behaviors, occurred in LMEs which share common productive features, despite different geographical locations and (2) demersal fisheries, which are more affected by economic drivers and tend to concentrate in LMEs in the Northern Hemisphere. Our analysis can be regarded as a first step towards the challenging scope of describing the relative influence of environmental and economic drivers on exploited ecosystems.