Stable Classifications Research Articles

Beyond central-tendency: If we agree discrete vegetation communities do not exist, should we investigate other methods of clustering?

Clustering is indispensable in the quest for robust vegetation classification schemes that aim to partition, summarise and communicate patterns. However, clustering solutions are sensitive to methods and data and are therefore unstable, a feature that is usually attributed to noise. Viewed through a central-tendency lens, noise is defined as the degree of departure from type, which is problematic since vegetation types are abstractions of continua, and so noise can only be quantified relative to the particular solution at hand. Graph theory models the structure of vegetation data based on the interconnectivity of samples. Through a graph-theoretic lens, the causes of instability can be quantified in absolute terms via the degree of connectivity among objects. We simulated incremental increases in sampling intensity in a dataset over five iterations and assessed classification stability across successive solutions derived using algorithms implementing, respectively, models of central-tendency and interconnectivity. We used logistic regression to model the likelihood of a sample changing groups between iterations as a function of distance to the centroid and degree of interconnectivity. Our results show that the degree to which samples are interconnected is a more powerful predictor of instability than the degree to which they deviate from their nearest centroid. The removal of weakly interconnected samples resulted in more stable classifications, although solutions with many clusters were apparently inherently less stable than those with few clusters, and improvements in stability flowing from the removal of outliers declined as the number of clusters increased. Our results reinforce the fact that clusters abstracted from continuous data are inherently unstable and that the quest for stable, fine-scale classifications from large regional datasets is illusory. Nevertheless, our results show that using models better suited to the analysis of continuous data may yield more stable classifications of the available data.

On tuning a mean-field model for semi-supervised classification

Semi-supervised learning (SSL) has become an interesting research area due to its capacity for learning in scenarios where both labeled and unlabeled data are available. In this work, we focus on the task of transduction—when the objective is to label all data presented to the learner—with a mean-field approximation to the Potts model. Aiming at this particular task we study how classification results depend on β and find that the optimal phase depends highly on the amount of labeled data available. In the same study, we also observe that more stable classifications regarding small fluctuations in β are related to configurations of high probability and propose a tuning approach based on such observation. This method relies on a novel parameter γ and we then evaluate two different values of the said quantity in comparison with classical methods in the field. This evaluation is conducted by changing the amount of labeled data available and the number of nearest neighbors in the similarity graph. Empirical results show that the tuning method is effective and allows NMF to outperform other approaches in datasets with fewer classes. In addition, one of the chosen values for γ also leads to results that are more resilient to changes in the number of neighbors, which might be of interest to practitioners in the field of SSL.

Placental characteristics of selective intrauterine growth restriction with changing patterns in umbilical artery Doppler flow in monochorionic diamniotic twins.

To compare the placental features between selective intrauterine growth restriction (sIUGR) patients, with dynamic classification (DC) and stable classifications (SC). A cohort study was conducted from 1st April 2013 to 1st April 2020, we recruited sIUGR patients who had delivered at our center and examined placental characteristics with dye injection. The primary outcome was placental characteristics. The prevalence of large artery-artery anastomosis (AAA) was significantly higher in sIUGR with DC than sIUGR with SC (88.2 vs. 46.6%, p=0.001). The total diameter of AAA was significantly larger in sIUGR with DC than sIUGR with SC [2.9 (1.4, 7.0) vs. 2.4 (0.3, 7.1)mm, p=0.032]. The total number of artery-vein anastomosis was significantly smaller in sIUGR with DC than sIUGR with SC [3(1,12) vs. 5 (1, 15), p=0.023]. The incidence of neonatal asphyxias of growth-restricted fetuses were higher in sIUGR with DC than sIUGR with SC (23.5 vs. 5.9%, p=0.033). Large AAA is probably associated to sIUGR with dynamic classification.

Long-term outcomes of distal locking in extracapsular fractures treated with trochanteric Gamma3 nails

BackgroundFew publications have assessed long-term results of distal locking of short endomedullary nails for extracapsular hip fracture. Virtually all of them focus on immediate differences. Criteria for the use of static or dynamic locking are unclear in most nailing systems, and use is advised in unstable fracture patterns or with risk of bell-clapper effect, but often influenced by the “orthopaedic school”.Materials and methodsThis is a historical cohort study on patients diagnosed and operated in 2014 and followed up until endpoint, considered as consolidation or major complication, plus evaluation of overall long-term survival. They were categorised as static distal locking (ST) or dynamic distal locking (DN). Both are comparable, except for all stable pre-operative classifications, Fracture Mobility Score (FMS) at discharge, and immediate post-operative loading, all of which were in favour of DN.ResultsConsolidation took place in > 95% of patients, with a non-statistically significant delay trend in ST. Less than 6% in both ST and DN had major complications, with no differences. Most cases suffered early cut-out. Significant fracture collapse was the most frequent minor complication. There were more statistically significant minor and total complications in ST. Infection, without differences, can precede cut-out. Lateral thigh pain was similar and could be related to back-out. In DN, 21.1% of cases were truly dynamised. We did not find differences in mobility or in long-term survival.ConclusionsAny type of distal locking seems to be safe for consolidation, despite a slightly longer consolidation time in static locking. Early cut-out was the main complication, while others were very infrequent, which is an advantage over helical blade devices. There was a higher rate of minor and overall mechanical complications in ST, but infection and lateral thigh pain were similar. Most non-traumatic mechanical complications occurred around 5–6 weeks. About one in five of the DN truly dynamised, with all cases occurring before 8 weeks. Mobility until endpoint and overall long-term survival were not influenced by the locking mode used.Level of evidenceTherapeutic study, level 2b.

Advances in systematics of bats: towards a consensus on species delimitation and classifications through integrative taxonomy

The order Chiroptera comprises all bat species and is the second-most diverse order of mammals. Organization of this diversity into well-delimited taxa embodying their evolutionary affinities remains challenging, particularly because the relationships among various groups at most taxonomic levels are not yet fully resolved. Historically, systematic questions were approached independently using tools and methodologies within researchers' individual expertise. With technological advances, this trend has shifted to more comprehensive analyses, merging independent datasets from various lines of research. Nevertheless, disagreements among phylogenetic relationships inferred from independent datasets, and derived from the complexity of modes of diversification and speciation, still preclude the achievement of a single, comprehensive, and stable classification framework for bats. In this context, integrative taxonomy uses complementarity among various disciplines to improve the outcome of hypothesis testing, which can be achieved by congruence or cumulation of datasets. Here, we provide a review of bat systematics, with emphasis on the instances where integrative taxonomy has proven to be crucial in delimitation of taxa. We focus on examples of true integration or integration using discrete data to support delimitation at different taxonomic levels. In addition, we discuss the difficulties associated with the integration of incongruent phylogenies, as well as how to incorporate information crucial to phylogenetic decisions from data not easily comparable to most common datasets. Our goal is to provide exemplar guidelines that can be useful to taxonomists for the creation of stable classifications without subjectivity and limitations of single datasets. Accordingly, we present a review of recent issues in chiropteran systematics, including comments on how the use of integrative taxonomy helped to solve them.

Joint optimization of cluster number and abundance transformation for obtaining effective vegetation classifications

AbstractQuestionIs it possible to determine which combination of cluster number and taxon abundance transformation would produce the most effective classification of vegetation data? What is the effect of changing cluster number and taxon abundance weighting (applied simultaneously) on the stability and biological interpretation of vegetation classifications?LocalityEurope, Western Australia, simulated data.MethodsReal data sets representing Hungarian sub‐montane grasslands, European wetlands, and Western Australian kwongan vegetation, as well as simulated data sets were used. The data sets were classified using the partitioning around medoids method. We generated classification solutions by gradually changing the transformation exponent applied to the species projected covers and the number of clusters. The effectiveness of each classification was assessed with a stability index. This index is based on bootstrap resampling of the original data set with subsequent elimination of duplicates. The vegetation types delimited by the most stable classification were compared with other classifications obtained at local maxima of the stability values. The effect of changing the transformation power exponent on the number of clusters, indexed according to their stability, was evaluated.ResultsThe optimal number of clusters varied with the power exponent in all cases, both with real and simulated data sets. With the real data sets, optimal cluster numbers obtained with different data transformations recovered interpretable biological patterns. Using the simulated data, the optima of stability values identified the simulated number of clusters correctly in most cases.ConclusionsWith changing the settings of data transformation and the number of clusters, classifications of different stability can be produced. Highly stable classifications can be obtained from different settings for cluster number and data transformation. Despite similarly high stability, such classifications may reveal contrasting biological patterns, thus suggesting different interpretations. We suggest testing a wide range of available combinations to find the parameters resulting in the most effective classifications.

CLASSIFICATION OF BENTHIC BIOCENOSES OF THE LOWLAND RIVER TUDOVKA (TVER REGION, RUSSIA) USING COMMUNITY FEATURES

Within the joint Russian-Austrian monitoring programme “REFCOND_VOLGA (2006 – 20XX)”, monitoring sites were established in the headwaters of the Volga (Tver Region). River Tudovka, a right tributary to the Volga River, was included within this monitoring programme as its catchment is partly protected and has only few anthropogenic activities. The monitoring activities include physico-chemical and hydraulic parameters as well as biota with a focus is on benthic organisms (diatoms and macrozoobenthos). In this work, the longitudinal patterns in community structure are classified in the lowland river Tudovka using a novel feature-based approach taken from signal processing theory. The method first clusters field sampling data into longitudinal classes (upper, middle, lower course). Community features based on the relative frequency of individual species occurring per class are then generated. We apply both generative and discriminative classification methods. The application of generative methods provides data models which predict the probability of a new sample to belong to an existing class. In contrast, discriminative approaches search for differences between classes and allocate new data accordingly. Leveraging both methods allows for the creation of stable classifications. On this basis we show how the community features can be used to predict the longitudinal class. The community features approach also allows for a robust cross-comparison of investigation reaches over time. In cases where suitable long-term data set are available, predictive models using this approach can also be developed.

Towards stable classifications

Towards stable classifications

Using simulations to investigate the longitudinal stability of alternative schemes for classifying and identifying children with reading disabilities

ABSTRACTThe present study employed data simulation techniques to investigate the 1-year stability of alternative classification schemes for identifying children with reading disabilities. Classification schemes investigated include low performance, unexpected low performance, dual-discrepancy, and a rudimentary form of constellation model of reading disabilities that included multiple criteria. Data from a previously published study were used to construct a growth model of reading development. The parameters estimated from this model were then used to construct three simulated data sets wherein the growth parameters were manipulated in one of three ways: a stable-growth pattern, a mastery learning pattern, and a fan-spread pattern. Results indicated that overall the constellation model provided the most stable classifications across all conditions of the simulation, and that classification schemes were most stable in the fan-spread condition and were the least stable under the mastery learning growth pattern. These results also demonstrate the utility of data simulations in reading research.

Evolution of Classification Systems

The question of how to order our knowledge is as old as systematic acquisition, circulation, and storage of knowledge. Classification systems are known since ancient times. Web technologies foster self-organized knowledge production and folksonomies are pictured as counter examples to expert-based designed knowledge ordering systems, such as library classifications or domain-specific ontologies. However, a closer look into the structure of user-generated content (e.g., the category system of Wikipedia) and its temporal evolution reveals surprising similarities to more traditional classification systems. In related work we have used evolutionary analysis of the UDC, treating it as a stable reference system over against the volatility of the knowledge landscape represented by the constantly shifting knowledge network in Wikipedia (Akdag Salah et al. 2011; Scharnhorst et al. 2011). We also have used the UDC as a case study in ontogeny to demonstrate the instantiating evolutionary tree of the UDC over time (Akdag Salah et al. 2012), reflecting the socio-cultural knowledge landscape of the 20th century in which it developed. We see KOSs functioning like artificial languages to describe information objects in a controlled way, rather than as hierarchical trees designed to allocate documents. In this manner both the user-generated category system of Wikipedia and stable reference classifications give evidence of gradual evolution of intension over time as lexical content mutates rather than sudden or jarring theoretical shifts in base knowledge. In this talk we feature work done to visualize the evolution of classification systems, to compare them and to develop new interfaces to collections that make use of available metadata, including classifications.

Sociometric research in pedagogy

The use of sociometric research in pedagogy implies consideration of methodological demands pertaining to the selection and formulation of sociometric criteria, the selection of sociometric techniques, instruments and sociometric group classification methods. Various authors agree that sociometric criteria are basic relevant activities of the group they refer to and that they should be meaningful, understandable to every group member and clearly and precisely formulated. The selection of the appropriate sociometric technique and instrument has to be adjusted with the goal of sociometric research, along with the consideration of the advantages and short?comings of different sociometric techniques and instruments. The existing research provides very divergent results on the adequacy of different classification methods in identifying stable sociometric groups. Still, the majority of studies have confirmed that the two-dimensional rating scale method yields more stable classifications than the classification methods based on peer nominations, as well as that it is methodologically more justifiable to use cluster analysis in identifying stable sociometric status groups than the traditional classification methods.

Phylogenetic Relationships of the Afro-Malagasy Members of the Large Genus Begonia Inferred from trnL Intron Sequences

The first extensive cpDNA phylogeny of African Begonia was estimated with chloroplast trnL intron DNA sequences. A total of 81 Begonia species was sampled and analyzed using parsimony and maximum likelihood methods. Analyses showed well-resolved sectional relationships, but little resolution at the base of the phylogeny. Malagasy species of Begonia formed a monophyletic group, as were the mainland African wingless fleshy-fruited species with the exception of section Mezierea. Species in section Mezierea are not as closely related to other wingless fleshy-fruited species as once thought. The section is polyphyletic, with the Malagasy members of the section nested within the Malagasy clade. The phylogenetic position of species in section Mezierea demonstrates that fleshy fruits have evolved at least twice in African Begonia. The terrestrial, 2-tepaled, West and Central African yellow-flowered begonias and their related species are also monophyletic, with the monotypic section Cristasemen and B. iucunda holding a solitary, basal position. The predominantly East and South African species in sections Rostrobegonia, Augustia, Sexalaria, and Peltaugustia show greater affinity with Asian and American, rather than other African begonias, and are interdigitated among them; therefore African Begonia are not monophyletic. The trnL intron is a useful region for identifying monophyletic groups across the whole genus and breaking up the genus into workable units. This demonstrates the value of reconstructing phylogenies of large genera in order to achieve more stable classifications.

Detecting Industry Substructure: Case of Banking, Steel and Pharmaceutical Industries

This paper further develops the stock return method proposed by Cho and McKelvey (1996). It is claimed that the method may detect industry substructure in an objective and effective way. Although they are statistically significantly different while avoiding any artifactual statistical results (Barney and Hoskisson, 1990; Peteraf and Shanley, 1997), the groups found in Cho and McKelvey (1996) fail to show clear face validity. Furthermore, the one-year sample window used in their study may be too short of a time to capture a sufficient number of outside disturbances that are the basis for grouping. In this paper, the sample period windows are extended from the previous one-year window to 6 different windows, namely 1-year, 2-year, 3-year, 5-year, 7-year and 9-year window spans. To test its stability in different time periods, clustering results of year 2000-04 are compared to those of year 1988-96. By formally applying different sample windows and time periods, the limitations of small window and unknown stability (Cho and McKelvey, 1996) are supposed to be dismantled. Furthermore, this paper examines whether daily returns or weekly returns are better to use in the stock return method. Among the largest firms in their market capitalization, 30 sample firms listed in New York Stock Exchange or NASDAQ (NYDAQ) are carefully chosen from the steel, banking, and pharmaceutical industries. We find that the stock return method produces stable group classifications consistently across different sample windows and time periods. In our particular sample, the groups found show clear face validity, say its correct industry membership. As the time span increases from 1 year to 9 years, the group structures become clearer and tighter. The grouping structure found in the 1988-96 time period has been consistently maintained in the 2000-04 time period. Daily returns produce the same grouping results with those from weekly returns, but they are better because they can detect groups sooner or with smaller time windows. Several conclusions are drawn from the study. First, the stock return method can effectively identify industry substructure as maintained in Cho and McKelvey (1996). The findings confirm that industry substructure can be reliably and validly separated, and that substructure stability has been longitudinally maintained across different sampling windows and time periods. Second, the identified group structure is not artifactual. The historically consistent results from our method using 'hard' market-equilibrium data render a high level of validity on our finding. Third, the findings are objective because the sample data used are 'hard' data, and the stock return method has no subjective decisions buried within it (including clustering methods).

Color classification by chimpanzees ( Pan troglodytes) in a matching-to-sample task

We investigated chimpanzees’ color classification using a matching-to-sample procedure. One of the two subjects had learned symbolic color names through long-term training, while the other had received less training and had a limited understanding of color names. The results showed similar distributions of classified colors in a color space, irrespective of the subjects’ differential color-naming experience. However, the chimpanzee with little color-naming experience showed less stable classifications. These results suggest common features of color classification in chimpanzees, as well as the influence of color experience and/or the learning of color names on the stability of classification of colors.

A Concern for Evidence and a Phylogenetic Hypothesis of Relationships among Epicrates (Boidae, Serpentes)

-Character congruence, the principle of using all the relevant data, and character independence are important concepts in phylogenetic inference, because they relate directly to the evidence on which hypotheses are based. Taxonomic congruence, which is agreement among patterns of taxonomic relationships, is less important, because its connection to the underlying character evidence is indirect and often imperfect. Also, taxonomic congruence is difficult to justify, because of the arbitrariness involved in choosing a consensus method and index with which to estimate agreement. High levels of character congruence were observed among 89 biochemical and morphological synapomorphies scored on 10 species of Epicrates. Such agreement is consistent with the phylogenetic interpretation attached to the resulting hypothesis, which is a consensus of two equally parsimonious cladograms: (cenchria (angulifer (striatus ((chrysogaster, exsul) (inornatus, subflavus) (gracilis (fordii, monensis)))))). Relatively little (11.4%) of the character incongruence was due to the disparity between the biochemical and morphological data Each of the clades in the consensus cladogram was confirmed by two or more unique and unreversed novelties, and six of the eight clades were corroborated by biochemical and morphological evidence. Such combinations of characters add confidence to the phylogenetic hypothesis, assuming the qualitatively different kinds of data are more likely to count as independent than are observations drawn from the same character system. Most of the incongruence occurred in the skeletal subset of characters, and much of that independent evolution seemed to be the result of paedomorphosis. [Biochemical data; character congruence; character independence; Epicrates; evidence; morphological data; paedomorphosis; phylogenetic systematics; taxonomic congruence; total evidence.] The goals of cladistics and phenetics are different. The former estimates phylogeny (Hennig, 1966), the latter seeks stability and convenience in classification (Sokal, 1986:424). Unfortunately, these objectives have become confused, especially where stability is concerned. For example, Hillis (1987:35) stated that [c]lassifications are best based on information in common among multiple data sets (i.e., consensus trees), whereas the best estimate of phylogeny and best estimate of character evolution are represented in the analysis of the combined data sets. The importance of a conservative information storage-retrieval system is undeniable (e.g., International Code of Zoological Nomenclature, 1985:3), and pheneticists sought justification for their methods in this simple truth (Sokal and Sneath, 1963). Even cladists attributed some significance to stability per se, because the lack of it confounds attempts to discover the single historical pattern (Schuh and Farris, 1981). Not unexpectedly, cladists responded (e.g., Farris, 1971) to the proposition that phenetic methods produce more stable classifications (e.g., Sokal and Sneath, 1963: 264). Analytical precision was demanded by the contestants, and an incredible array of consensus methods and indexes were developed (see lists below). It is important to bear in mind that only information on taxonomic grouping can be used to evaluate the pheneticists' claims, because character congruence is difficult to judge in phenetic analyses. Eventually, the controversy over stability became a relative issue, with cladists arguing greater stability for classifications produced by their methods, as if stability was a goal of phylogenetic systematics. Further, Nelson and Platnick (1981:219; see also Nelson, 1979) attempted to justify cla-

Factors affecting the character stability of classifications

The effect of six resemblance coefficients (taxonomic distance, Manhattan distance, correlations, cosines, and two new general dissimilarity coefficients) on the character stability of classifications based on six data sets was evaluated. The six data sets represent a variety of organisms, and of ratios of number of characters to number of OTUs, and were randomly bipartitioned 100 times. The results of matrix correlations, cophenetic correlations and two consensus measures indicate that no one resemblance coefficient is uniformly better than all others when evaluated in terms of the stability of a classification, although taxonomic distance and Manhattan distance produce relatively more stable classifications than the other resemblance coefficients. An index of dimensionality, the stemminess and cophenetic correlations of classifications were calculated for the six data sets and also for 20 data sets analyzed in an earlier study. Regression analysis of stability on the ratio of number of characters to the number of OTUs, dimensionality, stemminess, and cophenetic correlations explained more than 70% of the variance in stability. Of the four factors, the ratio was by far the most important. Stemminess and dimensionality contributed little when considered singly, and did not add appreciably to the variance explained by ratio and cophenetic correlations.

The Use of Trend-Surface Parameters in Inter-Urban Comparisons

The general problem of comparing isarithmic maps of different urban areas is examined. Proposed solutions by Merriam and Sneath (1966) and Haggett (1967) are revised in the light of the indicated instability of coefficients under shifts in the grid systems for such maps. A number of stable classifications of city structures can be built up which are invariant under orthogonal grid rotations but most of these break down under non-orthogonal rotations. Variances accounted for by successive terms (linear, quadratic, cubic) appeared stable and are recommended for preliminary analysis. The fundamental problems of pattern analysis remain unresolved by trend-surface coefficients. The argument in the paper is largely empirical and is illustrated from simulated urban patterns and by fifteen sample metropolitan areas from the United States.