Self-organizing Maps as Substitutes for K-Means Clustering

Bação, Fernando; Lobo, Victor; Paìnho, Marco

doi:10.1007/11428862_65

Cited by 189 publications

(62 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Los SOM son relativamente insensibles a los valores perdidos, tolerando a la vez datos con una distribución no normal ); esto le permite prescindir de verificaciones de difícil cumplimiento, haciéndolo válido para cualquier distribución de datos. Por otro lado como método de clusterización el SOM es más robusto que por ejemplo el K-means, aunque en su contra requiere mayor tiempo de computación (Bação, Lobo, & Painho, 2005;Gomes et al, 2007).…”

Section: Metodología De Investigaciónunclassified

Methodology of Decision Support through GIS and Artificial Intelligence: Implementation for Demographic Characterization of Andalusia based on Dwelling

Abarca-Álvarez¹,

Campos-Sánchez²,

Reinoso-Bellido³

2017

est

View full text Add to dashboard Cite

Resumenos Sistemas de Información Geográfica (SIG) han sido ampliamente utilizados para el almacenamiento y gestión de la información territorial, mostrándose especialmente útiles para el análisis y para la verificación de hipótesis previamente formuladas y con componentes espaciales relevantes. Existen metodologías heurísticas que en contextos como los actuales, de sobre-abundancia de datos, permiten evidenciar sus coherencias, sin requerir necesariamente hipótesis o formulaciones previas para generar conocimiento. Se propone el uso combinado de (i) técnicas procedentes de la Inteligencia Artificial, como son las Redes Neuronales Artificiales (ANN) del tipo Mapa Auto-organizado (SOM), que han demostrado ser muy eficaces y robustas clasificando y caracterizando perfiles en los datos; integradas con (ii) técnicas de Machine Learning como son los árboles de decisión, singularmente funcionales en la creación de modelos predictivos e interpretables para formular hipótesis explicativas de los perfiles anteriores a partir de otras variables diferenciadas. La investigación plantea combinar SIG, SOM y árboles de decisión para la construcción de modelos explicativos de los perfiles demográficos y sociales de Andalucía, a partir de datos de bajo coste sobre la dimensión residencial. Se verifica la viabilidad de tales modelos predictivos y su alto valor para la comprensión y para la toma de decisiones sobre tales territorios.Palabras clave: árbol de decisión SIG, DSS, mapa auto-organizado. Abstract:Geographic Information Systems (GIS) have been widely used for the storage and management of territorial information, being especially useful for the analysis and verification of previously formulated hypotheses and coexisting with relevant spatial components. There are heuristic methodologies that, in contexts such as the present one, of data over-abundance, allow showing their coherence, not necessarily requiring hypotheses or previous formulations to generate knowledge. The combined use of (i) Artificial Intelligence techniques such as the Artificial Neural Network (ANN), namely the Self-Organized Maps (SOM), is proposed. They are very effective and robust by classifying and characterizing profiles in the data. They interact with (ii) machine learning techniques such as decision trees, which are singularly functional in the creation of predictive and interpretable models, with the intention of formulating explanatory hypotheses of the previous profiles, working with other different variables. The research proposes the combination of GIS, SOM and decision trees for the construction of explanatory models of the demographic and social profiles of Andalusia, based on low cost data on the residential dimension. The feasibility of such predictive models and their great value for understanding and as decision support on such territories are evaluated satisfactorily.

show abstract

Section: Metodología De Investigaciónunclassified

Methodology of Decision Support through GIS and Artificial Intelligence: Implementation for Demographic Characterization of Andalusia based on Dwelling

Abarca-Álvarez¹,

Campos-Sánchez²,

Reinoso-Bellido³

2017

est

View full text Add to dashboard Cite

show abstract

“…A SOM with a number of units equal to the number of clusters in the data set and a neighborhood equal to zero will act as a traditional clustering technique (Kaski, 1997). A SOM may, however, be used in two very distinct ways: a large SOM, also known as emergent SOM, with many units, used for exploratory data analysis and cluster detection (Ultsch and Herrmann, 2005), and a small SOM for cluster centroid determination (Bação et al, 2005b). In this study, both SOM and GEO3DSOM are used for exploratory detection of clusters.…”

Section: Standard Sommentioning

confidence: 99%

“…In emergent SOM's on the other hand, clearly separated groups of units may or may not be detected. Small SOM's used for centroid determination will act as a robust K-means initialization in the first training iterations and due to the decrease of learning radius and neighborhood during training, the SOM will perform exactly as a K-means clustering in the final steps of the learning process (Bação et al, 2005b). Compared to K-means clustering and fuzzy c-means clustering, SOM has, in addition to the ability of SOM to directly visualize the results of the clusters in terms of the original variables, the advantage that the number of clusters does not need to be specified a priori.…”

Section: Standard Sommentioning

confidence: 99%

“…The advantage of K-means clustering and fuzzy c-means clustering on the other hand, is the ease of implementation since there are less parameters to be chosen. On the performance of clustering of SOM compared to other techniques a debate still exists in literature (overviews can be found in Bação et al, 2005b, andMingoti andLima, 2006). Provided the SOM is parameterized correctly, SOM will outperform K-means clustering since SOM is less sensitive to local optima compared to K-means Bação et al (2005b).…”

Section: Standard Sommentioning

confidence: 99%

See 1 more Smart Citation

Exploratory data analysis and clustering of multivariate spatial hydrogeological data by means of GEO3DSOM, a variant of Kohonen's Self-Organizing Map

Peeters

Bação

Lobo

et al. 2007

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

Abstract. The use of unsupervised artificial neural network techniques like the self-organizing map (SOM) algorithm has proven to be a useful tool in exploratory data analysis and clustering of multivariate data sets. In this study a variant of the SOM-algorithm is proposed, the GEO3DSOM, capable of explicitly incorporating three-dimensional spatial knowledge into the algorithm. The performance of the GEO3DSOM is compared to the performance of the standard SOM in analyzing an artificial data set and a hydrochemical data set. The hydrochemical data set consists of 131 groundwater samples collected in two detritic, phreatic, Cenozoic aquifers in Central Belgium. Both techniques succeed very well in providing more insight in the groundwater quality data set, visualizing the relationships between variables, highlighting the main differences between groups of samples and pointing out anomalous wells and well screens. The GEO3DSOM however has the advantage to provide an increased resolution while still maintaining a good generalization of the data set.

show abstract

“…Although there are several different variations and optimizations of K-Means algorithm 21 , this paper is focused on its four methods (Lloyd, Forgy, MacQueen and Hartigan-Wong).…”

Section: Clustering Techniquesmentioning

confidence: 99%

A Benchmarking Study of Clustering Techniques Applied to a Set of Characteristics of MOOC Participants

Cabedo¹,

Edmundo²,

Castro³

2016 ASEE Annual Conference &Amp; Exposition Proceedings

View full text Add to dashboard Cite

Rosa Cabedo is Engineer in Computer Science and currently Ph.D. Student at Technical University of Madrid (Spain) in the field of Open Education. The final purpose of her research is the identification and analysis of the profiles of language MOOC participants and the features of language learning (interaction, feedback, evaluation, certification) in order to adequate the design to MOOC format to facilitate the linguistic and communicativa competences improvement of LMOOC participants and their professional development. Her research focuses on the analysis of the heterogeneity of (L)MOOCs participants with the help of clustering techniques. ) format is characterized by the great diversity of enrolled people. Moreover, the lack of prior knowledge of their profiles constitutes an important barrier with a view to identifying and getting a better understanding of underlying relationships in the internal structure of the features that make up the profile of the participants in those courses. This paper has the aim of identifying and analyzing the feasible set of MOOC participants' profiles by running two unsupervised clustering techniques, KMeans as a partitional clustering algorithm and Kohonen's Self-Organizing Maps (SOMs), hereinafter SOM, as a representative technique of Artificial Neural Networks (ANNs).The selected dataset for this paper comes from the MOOCKnowledge project data collection, which provides an opportunity to work with real-world data from hundreds of people. KMeans and SOM algorithms are performed with a subset of participants' features as input data. The clustering evaluation, meanwhile, is achieved with a selection of indices, an intracluster measure and an overall quality criterion for K-Means, and two measures related to topological ordering for SOM.The comparison of internal structure of both clustering (set of profiles) shows that there are similarities between them on the one hand and some pinpointed differences that can not be evaluated in advance without the opinion of an expert familiarized with the specifications of the MOOC on the other.Therefore, this comparison can not be considered conclusive until after a preliminary study of the results of the clustering interpretation for both algorithms. Finally, although it is not determined the clustering that best fits between K-Means and SOM, this study might help to provide a methodological guide on how to identify and select the appropriate clustering according to several quality criteria. Key WordsMOOC profiles, K-Means, Kohonen's Self-Organizing Maps, SOM, cluster analysis, clustering IntroductionThis paper has the final purpose of dealing with a comparative study of two different clustering approaches (K-Means and SOM) on a selected set of participants' features of a MOOC in the scope of the personal development. With this study, clustering could be discovered as a useful exploratory technique for identifying and analyzing MOOC participants' profiles, a format characterized by the great diversity of enrolled people. The heterogeneity...

show abstract

Self-organizing Maps as Substitutes for K-Means Clustering

Cited by 189 publications

References 25 publications

Methodology of Decision Support through GIS and Artificial Intelligence: Implementation for Demographic Characterization of Andalusia based on Dwelling

Methodology of Decision Support through GIS and Artificial Intelligence: Implementation for Demographic Characterization of Andalusia based on Dwelling

Exploratory data analysis and clustering of multivariate spatial hydrogeological data by means of GEO3DSOM, a variant of Kohonen's Self-Organizing Map

A Benchmarking Study of Clustering Techniques Applied to a Set of Characteristics of MOOC Participants

Contact Info

Product

Resources

About