Cluster Analysis Of Time Series Research Articles

Integration of data analysis methods in syndromic surveillance systems.

Syndromic surveillance systems perform real-time analysis of health data to enable early identification of potential public health threats, evaluating whether distributional parameters have been increased beyond a threshold. This paper presents the applied data analysis methods in five non-industrial surveillance systems. Four time series and spatial cluster analysis methods were found to be implemented: SMART, EWMA, CuSum and WSARE. Combined use both spatial and time series methods is found in the presented surveillance applications. Data analysis methods for syndromic surveillance are a constantly emerging field, while new statistical methods and algorithms are implemented into surveillance systems.

Muscle transcriptome profiling in divergent phenotype swine breeds during growth using microarray and RT‐PCR tools

Using an array consisting of 10 665 70-mer oligonucleotide probes, the longissimus dorsi muscle tissue expression during growth in nine pigs belonging to Casertana (CT), an autochthonous breed characterized by slow growth and a massive accumulation of backfat, was compared with that of two cosmopolitan breeds, Large White (LW) and a crossbreed (CB; Duroc × Landrace × Large White). The results were validated by real-time PCR. All animals were of the same age and were raised under the same environmental conditions. Muscle tissues were collected at 3, 6, 9 and 11 months of age, and a total of 173 genes showed significant differential expression between CT and the cosmopolitan genetic types at 3 months of age. Time series cluster analysis indicated that the CT breed had a different pattern of gene expression compared with that of the LW and the CB. Four of the eight clusters highlighted the gene differences between CT and the other two breeds, which were further supported by statistical analyses: clusters 4 and 5 contained a total of 71 genes that were underexpressed at 3 months of age, and cluster 3 and cluster 7 included 28 and 42 genes respectively that were overexpressed at 3 months of age. As expected, differentially expressed genes belonged to the category of genes coding for contractile fibres and transcription factors involved in muscle development and differentiation. These findings highlight muscle expression genes during pig growth and are useful to understand the genetic meaning of the different developmental rates.

Fuzzy clustering of time series in the frequency domain

Traditional and fuzzy cluster analyses are applicable to variables whose values are uncorrelated. Hence, in order to cluster time series data which are usually serially correlated, one needs to extract features from the time series, the values of which are uncorrelated. The periodogram which is an estimator of the spectral density function of a time series is a feature that can be used in the cluster analysis of time series because its ordinates are uncorrelated. Additionally, the normalized periodogram and the logarithm of the normalized periodogram are also features that can be used. In this paper, we consider a fuzzy clustering approach for time series based on the estimated cepstrum. The cepstrum is the spectrum of the logarithm of the spectral density function. We show in our simulation studies for the typical generating processes that have been considered, fuzzy clustering based on the cepstral coefficients performs very well compared to when it is based on other features.

A novel two-level clustering method for time series data analysis

Clustering analysis has been applied in a wild variety of fields such as biology, medicine, economics, etc. For time series clustering, dimension reduction methods like data sampling or piecewise aggregate approximation (PAA) algorithm are often applied to reduce data dimension before clustering. Consequently, the information of subsequence may be overlooked. Nevertheless, some properties of time series with the same sampling data may result in different clustering results after considering the subsequence information. In this paper, we propose a novel two-level clustering method named 2LTSC (two-level time series clustering), which considers both the whole time series, denoted as level-1 in the first level, and the subsequence information of time series, denoted as level-2 in the second level. The data length of level-2 could be different and thus is also considered in the second level in the proposed 2LTSC method. Through experimental evaluation, it is shown that the proposed two-level clustering method, which considers two different time granules at the same time, can provide different and deeper viewpoints for time series clustering analysis.

Evolution of transcriptome profiles during muscle development in Casertana and cosmopolite pig breeds

In order to determine candidate genes involved in production traits, the mRNA levels of muscle tissue of extremely different pig breeds (genotypes) were studied using microarray tool. Casertana, an autochthonous breed characterized by slow growth and a massive accumulation of backfat, was compared to Large White and to crossbred (Duroc x Landrace x Large White)pigs. The differential expression of muscle genes was evaluated on 3 pigs for each genetic type using a microarray consisting of 10,665 oligos. Animals were of the same age and raised in the same environmental conditions. Muscle tissues were collected at 3, 6, 9, and 11 months and a total of 219 (157 genes in the two main clusters) genes showed differential expression between genetic types. Time series cluster analysis indicated that Casertana breed had a different pattern of gene expressions compared to the Large White and the crossbreed. For Casertana pigs, a first cluster showed 105 genes under expressed at 3 months of age and a second cluster indicated 52 genes over expressed at 3 months of age, in comparison to the other genetic types. As expected, some of the differentially expressed genes were in the category of “contractile fiber” and transcriptional factors involved in muscle development and differentiation.

Cluster analysis of resting-state fMRI time series

Functional MRI (fMRI) has become one of the leading methods for brain mapping in neuroscience. Recent advances in fMRI analysis were used to define the default state of brain activity, functional connectivity and basal activity. Basal activity measured with fMRI raised tremendous interest among neuroscientists since synchronized brain activity pattern could be retrieved while the subject rests (resting state fMRI). During recent years, a few signal processing schemes have been suggested to analyze the resting state blood oxygenation level dependent (BOLD) signal from simple correlations to spectral decomposition. In most of these analysis schemes, the question asked was which brain areas “behave” in the time domain similarly to a pre-specified ROI. In this work we applied short time frequency analysis and clustering to study the spatial signal characteristics of resting state fMRI time series. Such analysis revealed that clusters of similar BOLD fluctuations are found in the cortex but also in the white matter and sub-cortical gray matter regions (thalamus). We found high similarities between the BOLD clusters and the neuro-anatomical appearance of brain regions. Additional analysis of the BOLD time series revealed a strong correlation between head movements and clustering quality. Experiments performed with T1-weighted time series also provided similar quality of clustering. These observations led us to the conclusion that non-functional contributions to the BOLD time series can also account for symmetric appearance of signal fluctuations. These contributions may include head motions, the underling microvasculature anatomy and cellular morphology.

A method for the detection of the severe rain‐on‐snow event on Banks Island, October 2003, using passive microwave remote sensing

Severe wintertime rain‐on‐snow (ROS) events create a strong ice layer (or layers) in the snow on arctic tundra that act as a barrier to ungulate grazing. They are linked with large‐scale ungulate (reindeer, caribou, elk, and musk‐ox) herd declines via starvation and reduced calf production rate when the animals are unable to penetrate the resulting subsnowpack ice layer. ROS events also produce considerable perturbation in the mean wintertime soil temperature under the snowpack. ROS is a sporadic but well‐known and significant phenomenon that is currently very poorly documented. Characterization of the distribution and occurrence of severe ROS events is based only on anecdotal evidence, indirect observations of carcasses found adjacent to iced snowpacks, and irregular detection by a sparse observational weather network. We have analyzed in detail a particular ROS event that took place on Banks Island in early October 2003 that resulted in the death of 20,000 musk oxen. We make use of multifrequency passive microwave imagery from the Special Sensor Microwave Imager satellite sensor suite in conjunction with a strong‐fluctuation‐theory (SFT) emissivity model. We show that a combination of time series analysis and cluster analysis based on microwave spectral gradients and polarization ratios provides a means to detect the stages of the ROS event resulting from the modification of the vertical structure of the snowpack, specifically wetting the snow, the accumulation of liquid water at the base of the snow during the rain event, and the subsequent modification of the snowpack after refreezing. SFT model analysis provides quantitative confirmation of our interpretation of the evolution of the microwave properties of the snowpack as a result of the ROS event. In addition to the grain coarsening owing to destructive metamorphism, we detect the presence of the internal water and ice layers, directly identifying the physical properties producing the hazardous conditions. This analysis offers the potential to characterize both the frequency and global distribution of ROS using multifrequency satellite passive microwave imagery.

Landscape level analysis of mountain pine beetle in British Columbia, Canada: spatiotemporal development and spatial synchrony within the present outbreak

Eruptive herbivores can exert profound landscape level influences. For example, the ongoing mountain pine beetle outbreak in British Columbia, Canada, has resulted in mortality of mature lodgepole pine over >7 million ha. Analysis of the spatio‐temporal pattern of spread can lend insights into the processes initiating and/or sustaining such phenomena. We present a landscape level analysis of the development of the current outbreak. Aerial survey assessments of tree mortality, projected onto discrete 12×12 km cells, were used as a proxy for insect population density. We examined whether the outbreak potentially originated from an epicenter and spread, or whether multiple localized populations erupted simultaneously at spatially disjunct locations. An aspatial cluster analysis of time series from 1990 to 2003 revealed four distinct time series patterns. Each time series demonstrated a general progression of increasing mountain pine beetle populations. Plotting the geographical locations of each temporal pattern revealed that the outbreak occurred first in an area of west‐central British Columbia, and then in an area to the east. The plot further revealed many localized infestations erupted in geographically disjunct areas, especially in the southern portion of the province. Autologistic regression analyses indicated a significant, positive association between areas where the outbreak first occurred and conservation lands. For example, the delineated area of west‐central British Columbia is comprised of three conservation parks and adjacent working forest. We further examined how population synchrony declines with distance at different population levels. Examination of the spatial dependence of temporal synchrony in population fluctuations during early, incipient years (i.e. 1990–1996) suggested that outbreaking mountain pine beetle populations are largely independent at scales >200 km during non‐epidemic periods. However, during epidemic years (i.e. 1999–2003), populations were clearly synchronous across the entire province, even at distances of up to 900 km. The epicentral pattern of population development can be used to identify and prioritize adjacent landscape units for both reactive and proactive management strategies intended to minimize mountain pine beetle impacts.

Cluster analysis of dynamic cerebral contrast-enhanced perfusion MRI time-series

We performed neural network clustering on dynamic contrast-enhanced perfusion magnetic resonance imaging time-series in patients with and without stroke. Minimal-free-energy vector quantization, self-organizing maps, and fuzzy c-means clustering enabled self-organized data-driven segmentation with respect to fine-grained differences of signal amplitude and dynamics, thus identifying asymmetries and local abnormalities of brain perfusion. We conclude that clustering is a useful extension to conventional perfusion parameter maps.

Clustering of financial time series with application to index and enhanced index tracking portfolio

A stochastic-optimization technique based on time series cluster analysis is described for index tracking and enhanced index tracking problems. Our methodology solves the problem in two steps, i.e., by first selecting a subset of stocks and then setting the weight of each stock as a result of an optimization process (asset allocation). Present formulation takes into account constraints on the number of stocks and on the fraction of capital invested in each of them, whilst not including transaction costs. Computational results based on clustering selection are compared to those of random techniques and show the importance of clustering in noise reduction and robust forecasting applications, in particular for enhanced index tracking.

Principal component cluster analysis of ECG time series based on Lyapunov exponent spectrum

In this paper we propose an approach of principal component cluster analysis based on Lyapunov exponent spectrum (LES) to analyze the ECG.time series. Analysis results of 22 sample-files of ECG.from the MIT-BIH database confirmed the validity of our approach. Another technique named improved teacher selecting student (TSS) algorithm is presented to analyze unknown samples by means of some known ones, which is of better accuracy. This technique combines the advantages of both statistical and nonlinear dynamical methods and is shown to be significant to the analysis of nonlinear ECG.time series.

Principal component cluster analysis of ECG time series based on Lyapunov exponent spectrum

Principal component cluster analysis of ECG time series based on Lyapunov exponent spectrum

Fuzzy cluster analysis of high-field functional MRI data

Functional magnetic resonance imaging (fMRI) based on blood–oxygen level dependent (BOLD) contrast today is an established brain research method and quickly gains acceptance for complementary clinical diagnosis. However, neither the basic mechanisms like coupling between neuronal activation and haemodynamic response are known exactly, nor can the various artifacts be predicted or controlled. Thus, modeling functional signal changes is non-trivial and exploratory data analysis (EDA) may be rather useful. In particular, identification and separation of artifacts as well as quantification of expected, i.e. stimulus correlated, and novel information on brain activity is important for both, new insights in neuroscience and future developments in functional MRI of the human brain. After an introduction on fuzzy clustering and very high-field fMRI we present several examples where fuzzy cluster analysis (FCA) of fMRI time series helps to identify and locally separate various artifacts. We also present and discuss applications and limitations of fuzzy cluster analysis in very high-field functional MRI: differentiate temporal patterns in MRI using (a) a test object with static and dynamic parts, (b) artifacts due to gross head motion artifacts. Using a synthetic fMRI data set we quantitatively examine the influences of relevant FCA parameters on clustering results in terms of receiver–operator characteristics (ROC) and compare them with a commonly used model-based correlation analysis (CA) approach. The application of FCA in analyzing in vivo fMRI data is shown for (a) a motor paradigm, (b) data from multi-echo imaging, and (c) a fMRI study using mental rotation of three-dimensional cubes. We found that differentiation of true “neural” from false “vascular” activation is possible based on echo time dependence and specific activation levels, as well as based on their signal time-course. Exploratory data analysis methods in general and fuzzy cluster analysis in particular may help to identify artifacts and add novel and unexpected information valuable for interpretation, classification and characterization of functional MRI data which can be used to design new data acquisition schemes, stimulus presentations, neuro(physio)logical paradigms, as well as to improve quantitative biophysical models.

Wind Climate Simulation over Complex Terrain and Wind Turbine Energy Output Estimation

A statistical-dynamical downscaling procedure is applied to investigate the climatological wind field over a complex terrain area in central Germany. The model domain, 80×87 km, is dominated by flat terrain in the westerly and northerly part and encompasses the Teuteburger Wald and the Wiehengebirge areas with hills up to 330 m a.m.s.l. in the southeasterly region. The downscaling procedure combines a large-scale regionally representative wind statistic and a high-resolution numerical atmospheric mesoscale model. A cluster analysis of a 12-years time series of radiosonde data provides 143 clusters each being a combination of the geostrophic wind components and the vertical temperature gradient. These parameter sets constitute the reference state for highly-resolved steady-state wind field simulations with a non-hydrostatic model. Weighting the resulting wind fields with the corresponding cluster frequency gives climatologically representative frequency distributions of the wind speed and -direction.

Explorative signal processing in functional MR imaging

Identification and separation of artifacts as well as quan- tification of expected, i.e., stimulus-correlated, and novel information on brain activity are important for both new insights in neuroscience and future developments in functional magnetic resonance imaging (MRI) of the human brain. Here, we present several examples in which gross head motion or physiologic motion (e.g., pulsation, respiration, large veins) could be identified and separated by using fuzzy cluster analysis of fMRI time series. Furthermore, our experience with single- and multislice fMRI (FLASH and EPI; 1.5 and 3 T) data analysis is summarized and several examples, including long echo time and high-resolution fMRI of the motor cortex, are discussed. Explorative signal processing in fMRI, based on fuzzy clustering, represents a robust and powerful tool for screening large fMRI data sets, extracting expected and novel functional activity of the human brain, and ob- taining improved reproducibility of fMRI results. Finally, it may help to improve or develop functional brain models which can then be tested by applying statistical models. © 1999 John Wiley & Sons, Inc. Int J Imaging

Using group seasonal indices in multi-item short-term forecasting

Methods for dealing with seasonal patterns of product sales can be categorized into two groups: those that forecast the demand for seasonal products by estimating the individual seasonal components for each product, and those that estimate the seasonal component by combining ‘similar’ products into a product line. An approach is proposed for the latter case, based on a synthesis of time series decomposition techniques and cluster analysis. Some initial experiments on a sample of retail sales data demonstrate its feasibility and give some comparative insights into this and alternative methods.

Attenuated Pulsatile Release of Prolactin in Men with Insulin-Dependent Diabetes Mellitus*

Pulsatile and circadian patterns of PRL release were studied in 11 insulin-dependent diabetic men by sampling blood every 10 min for 24 h and comparing the results to those obtained in 12 normal nondiabetic men. The diabetic men had a mean (+/- SE) 24-h serum PRL concentration of 5.5 +/- 0.42 micrograms/L, which was significantly lower than that in the nondiabetic men (9.3 +/- 0.86; P = 0.0008). Quantitative Cluster analysis of pulsatile PRL time series revealed a normal pulse frequency, but decreased maximal peak amplitude (6.6 +/- 0.5 vs. 11.8 +/- 1.1 micrograms/L; P = 0.0009), peak increment (2.6 +/- 0.24 vs. 4.0 +/- 0.3 micrograms/L; P = 0.009), peak area (126 +/- 15 vs. 192 +/- 19 micrograms/L.min; P = 0.03), and interpulse valley mean concentration (4.8 +/- 0.4 vs. 8.6 +/- 1.2 micrograms/L; P = 0.0007). PRL pulse incremental amplitude correlated significantly (r2 = 0.577; P = 0.007) and negatively with duration of disease. Fourier analysis disclosed a normal circadian rhythm of PRL release in diabetic men, with a mean circadian amplitude of 1.5 micrograms/L +/- 0.31, which peaked at 0201 h +/- 89 min (+/- SE). In summary, we have demonstrated significantly reduced mean 24-h serum PRL concentrations in men with poorly controlled insulin-dependent diabetes mellitus. The concomitant suppression of spontaneous PRL pulse amplitude, peak increment, and interpulse valley mean concentrations in the presence of normal pulse frequency is consistent with a reduced mass of PRL secreted per burst and/or accelerated metabolic clearance of PRL in men with type I diabetes mellitus.

Dynamics of 24-Hour Endogenous Cortisol Secretion and Clearance in Primary Hypothyroidism Assessed before and after Partial Thyroid Hormone Replacement*

Although various abnormalities of hypothalamic pituitary adrenal function have been reported in primary hypothyroidism, neither 24-h patterns of pulsatile cortisol release nor estimation of its endogenous secretion and clearance rates have been fully investigated in this clinical setting. We studied pulsatile and circadian patterns of cortisol secretion in six hypothyroid men [mean free T4 index, 0.59 +/- 0.22 (+/- SE); mean TSH, greater than 50 mU/L] by sampling blood at 20-min intervals for 24 h before (unreplaced) and then after 5-7 months of partial replacement treatment with levo-T4. Compared to a normal group, hypothyroid men had significantly elevated 24-h mean serum concentrations of cortisol (419 vs. 254 nmol/L; P less than 0.001), with no change in serum cortisol-binding globulin concentrations. Cluster analysis of cortisol time series revealed a normal pulse frequency, with significant increases in mean peak amplitude (527 vs. 331 nmol/L; P = 0.001), mean interpulse valley concentrations (384 vs. 204 nmol/L; P less than 0.05), and mean prepeak nadir concentrations (298 vs. 166 nmol/L; P less than 0.05). Cosinor analysis showed preserved circadian rhythmicity (i.e. normal mean circadian amplitude of cortisol release) in hypothyroidism, with a significant delay in the timing of circadian acrophases and an increase in the mesor (mean). Analysis of data by a multiple parameter deconvolution method demonstrated a normal 24-h endogenous cortisol production rate in the presence of significantly prolonged subject-specific half-life of cortisol disappearance (155 vs. 73 min; P less than 0.05). Partial replacement therapy with levo-T4 caused significant decreases in 1) mean 24-h serum cortisol concentrations (419 vs. 323 nmol/L; P less than 0.05); 2) mean cortisol peak amplitudes (527 vs. 375 nmol/L; P less than 0.05); 3) mean prepeak nadir concentrations (298 vs. 221 nmol/L; P less than 0.05); and 4) mean half-life of cortisol disappearance (155 vs. 112 min; P less than 0.0019). In summary, the present study of cortisol secretory dynamics in hypothyroid men has shown elevated mean 24-h serum concentrations of cortisol with preserved circadian rhymicity and normal endogenous production rates, but prolonged half-lives of cortisol disappearance. In conjunction with normal serum cortisol-binding globulin concentrations, these largely reversible findings suggest that significant hypercortisolemia in primary hypothyroidism is primarily due to decreased metabolic clearance of cortisol and a presumptive decrease in the negative feedback effect of cortisol on the hypothalamo-pituitary axis.

授業のカテゴリー分析における行動系列の抽出

An instruction, which sometimes is called teaching-learning process, is a communication process between teacher's teaching behaviors and learner's learning behaviors, and it can be represented as time series category data of teaching-learning behaviors. This paper describes a method for clustering some behavior sequences and also describes results applied on actual instruction. The method is constructed as follows. Partition of time series category data Cluster analysis for partitioned sequences Time series cluster analysis for some similar partitioned sequences As results can be shown as graph-representation, it is understandable and seems to be useful instructional study.

Cluster analysis of respiratory time series.

We have investigated the respiratory control system with the hypothesis that, although many variables such as minute ventilation (VI), tidal volume (VT), breathing period (TT), inspiratory duration (TI), and expiratory duration (TE) may be observed, the controller functions more simply by manipulating only 2 or 3 of these. Thus, if tidal volume is the only independent variable, TI being determined by the "off-switch" threshold, these variables should have very similar time courses. Anesthetized dogs were subjected to CO2 breathing and carotid sinus perfusion to stimulate both chemoreceptors. The time series of the variables VI, VT, TT, TE, and TI as well as PACO2 were determined on a breath by breath basis. Derived characteristics of these time series were compared using Cluster Analysis and the latent dimensionality of respiratory control determined by Factor Analysis. The characteristics of the time series clustered into 4 groups: magnitude (of the response), speed, variability and relative change. One respiratory factor accounted for 86% of the variance for the variability characteristics, 2 factors for magnitude (91%) and relative change (85%) and 3 factors for speed (89%). The respiratory variables were analysed for each of the 4 groups of characteristics with the following results: VT and TI clustered together only for the magnitude and relative change characteristics where as TT and TE clustered closely for all four characteristics. One latent factor was associated with the [TT-TE] group and the other usually with PACO2.