Large Time Series Datasets Research Articles

An evolutionary approach for efficient prototyping of large time series datasets

We here describe an algorithm based on an evolutionary strategy to find the prototype series of a set of time series, and we use Dynamic Time Warping (DTW) as a distance measure between series, and do not restrict the search space to the series in the set. The problem of calculating the centroid of a set of time series can be addressed as a minimization problem, using genetic algorithms. Our proposal may be considered among the set of non-classical approaches to genetic algorithms, where an individual gene is a candidate time series for being the centroid or representative of the whole set of series. The representation and operators of genetic algorithms are redesigned, in order to generate efficient summaries, the fitness function of each candidate series to be a prototype is approximated, comparing them only with a subset of randomly selected time series from the original dataset. Three areas are looked at in order to assess the goodness of our proposal: the performance of the prototype generated in terms of a fitness function, the consistency of the prototype generation for use in classical grouping algorithms, and its use in classification algorithms based on the nearest prototypes.

Mortality risk and burden associated with temperature variability in China, United Kingdom and United States: Comparative analysis of daily and hourly exposure metrics

BackgroundTemperature variability (TV) is closely associated with climate change, but there is no unified TV definition worldwide. Two novel composite TV indexes were developed recently by calculating the standard deviations of several days’ daily maximum and minimum temperatures (TVdaily), or hourly mean temperatures (TVhourly). ObjectivesThis study aimed to compare the mortality risks and burden associated with TVdaily and TVhourly using large time-series datasets collected from multiple locations in China, United Kingdom and United States. MethodsWe collected daily mortality and hourly temperature data through 1987 to 2012 from 63 locations in China (8 communities, 2006–2012), United Kingdom (10 regions, 1990–2012), and USA (45 cities, 1987–2000). TV-mortality associations were investigated using a three-stage analytic approach separately for China, UK, and USA. First, we applied a time-series regression for each location to derive location-specific TV-mortality curves. A second-stage meta-analysis was then performed to pool these estimated associations for each country. Finally, we calculated mortality fraction attributable to TV based on above-described location-specific and pooled estimates. ResultsOur dataset totally consisted of 23, 089, 328 all-cause death cases, including 93, 750 from China, 7,573,716 from UK and 15, 421, 862 from USA, respectively. In despite of a relatively wide uncertainty in China, approximately linear relationships were consistently identified for TVdaily and TVhourly. In the three countries, generally similar lag patterns of TV effects were consistently observed for TVdaily and TVhourly. A 1 °C rise in TVdaily and TVhourly at lag 0–7 days was associated with mortality increases of 0.93% (95% confidence interval [CI]: 0.12, 1.74) and 0.97% (0.18, 1.77) in China, 0.33% (0.15, 0.51) and 0.41% (0.21, 0.60) in UK, and 0.55% (0.41, 0.70) and 0.51% (0.35, 0.66) in USA, respectively. Larger attributable fractions were estimated using TVdaily than those using TVhourly, with estimates at 0–10 days of 3.69% (0.51, 6.75) vs. 2.59% (0.10, 5.01) in China, 1.14% (0.54, 1.74) vs. 0.98% (0.55, 1.42) in UK, and 2.57% (1.97, 3.16) vs. 1.67% (1.15, 2.18) in USA, respectively. Our meta-regression analyses indicated higher vulnerability to TV-induced mortality risks in warmer locations. ConclusionsOur study added multi-country evidence for increased mortality risk associated with short-term exposure to large temperature variability. Daily and hourly TV exposure metrics produced generally comparable risk effects, but the attributable mortality burden tended to be higher using TVdaily instead of TVhourly.

Deep Convolutional Neural Networks and Learning ECG Features for Screening Paroxysmal Atrial Fibrillation Patients

In this paper, a novel computationally intelligent-based electrocardiogram (ECG) signal classification methodology using a deep learning (DL) machine is developed. The focus is on patient screening and identifying patients with paroxysmal atrial fibrillation (PAF), which represents a life threatening cardiac arrhythmia. The proposed approach operates with a large volume of raw ECG time-series data as inputs to a deep convolutional neural networks (CNN). It autonomously learns representative and key features of the PAF to be used by a classification module. The features are therefore learned directly from the large time domain ECG signals by using a CNN with one fully connected layer. The learned features can effectively replace the traditional ad hoc and time-consuming user's hand-crafted features. Our experimental results verify and validate the effectiveness and capabilities of the learned features for PAF patient screening. The main advantages of our proposed approach are to simplify the feature extraction process corresponding to different cardiac arrhythmias and to remove the need for using a human expert to define appropriate and critical features working with a large time-series data set. The extensive simulations and case studies conducted indicate that combining the learned features with other classifiers will significantly improve the performance of the patient screening system as compared to an end-to-end CNN classifier. The effectiveness and capabilities of our proposed ECG DL classification machine is demonstrated and quantitative comparisons with several conventional machine learning classifiers are also provided.

A comprehensive graphical modeling platform designed for integrated hydrological simulation

This study developed a comprehensive graphical data processing and modeling system named Visual HEIFLOW (VHF) for integrated surface water-groundwater modeling. Its distinctive features include the following. First, VHF uses a generic multivariable-space-time data cube model, which enables the system to efficiently handle large time-series datasets over a large spatial domain. Second, VHF streamlines the entire modeling procedure from data preparation at the very beginning to visualization and analysis of modeling results in a uniform environment. Third, VHF allows updating the land use input at user-specified time points without manual intervention and therefore allows a direct simulation of the hydrological effects of changing land use. The applicability and versatility of VHF were demonstrated in a case study in the Heihe River Basin, the second largest inland river basin in China. The case study also demonstrated that VFH facilitates process understanding and supports management decision-making.

A Parallel and Optimization Approach for Land-Surface Temperature Retrieval on a Windows-Based PC Cluster

Land-surface temperature (LST) is a very important parameter in the geosciences. Conventional LST retrieval is based on large-scale remote-sensing (RS) images where split-window algorithms are usually employed via a traditional stand-alone method. When using the environment to visualize images (ENVI) software to carry out LST retrieval of large time-series datasets of infrared RS images, the processing time taken for traditional stand-alone servers becomes untenable. To address this shortcoming, cluster-based parallel computing is an ideal solution. However, traditional parallel computing is mostly based on the Linux environment, while the LST algorithm developed within the ENVI interactive data language (IDL) can only be run in the Windows environment in our project. To address this problem, we combine the characteristics of LST algorithms with parallel computing, and propose the design and implementation of a parallel LST retrieval algorithm using the message-passing interface (MPI) parallel-programming model on a Windows-based PC cluster platform. Furthermore, we present our solutions to the problems associated with performance bottlenecks and fault tolerance during the deployment stage. Our results show that, by improving the parallel environment of the storage system and network, one can effectively solve the stability issues of the parallel environment for large-scale RS data processing.

Effects of Promotions on Location-Based Social Media: Evidence from Foursquare

Location has been identified as a critical factor for the success of a business. For example, businesses in dense urban areas are exposed to more customers than businesses in sparsely populated neighborhoods, while proximity to a popular landmark can increase a business’s reach. This creates significant challenges for new businesses to expand their reach and customer base. However, the advancement of mobile, social, and spatial computing has led to the transformation of traditional web-based yellow pages to a mobile format (e.g., location-based social networks; LBSNs). This has allowed businesses that are not in prime locations to become visible to nearby customers. Furthermore, these platforms offer mechanisms that can serve as an affordable advertisement channel to local businesses. Specifically, a business can use LBSNs to promote special offers to customers who connect through the platform. Despite the promising anecdotal evidence, a systematic study of the effectiveness of this LBSN advertising paradigm has not yet been conducted. Using a large time-series data set of approximately 14 million venues in Foursquare, the largest LBSN to date, our work is the first to formally examine the effects of promotions through the platform. Our contribution is twofold. First, we identify no significant and robust evidence that can support the hypothesis of effective promotions for the specific platform. In particular, our main finding is that the probability of observing an increase in daily check-ins or new daily customers to a venue is rarely altered by the presence of a Foursquare promotion. Second, this finding motivates us to design a model that can predict the success of a promotion according to various relevant features. Our model can be used to inform the process of designing and launching successful promotions by evaluating the potential of candidate promotions before their actual release. The practical value of such prelaunch evaluations is elevated by the apparent scarcity of successful promotions, as revealed by our analysis.

Variable Selection in Time Series Forecasting Using Random Forests

Time series forecasting using machine learning algorithms has gained popularity recently. Random forest is a machine learning algorithm implemented in time series forecasting; however, most of its forecasting properties have remained unexplored. Here we focus on assessing the performance of random forests in one-step forecasting using two large datasets of short time series with the aim to suggest an optimal set of predictor variables. Furthermore, we compare its performance to benchmarking methods. The first dataset is composed by 16,000 simulated time series from a variety of Autoregressive Fractionally Integrated Moving Average (ARFIMA) models. The second dataset consists of 135 mean annual temperature time series. The highest predictive performance of RF is observed when using a low number of recent lagged predictor variables. This outcome could be useful in relevant future applications, with the prospect to achieve higher predictive accuracy.

A novel clustering-based method for time series motif discovery under time warping measure

The problem of time series motif discovery has attracted a lot of attention and is useful in many real-world applications. However, most of the proposed methods so far use Euclidean distance to deal with this problem. There has been one proposed method, called MDTW_WedgeTree, for time series motif discovery under DTW distance. But this method aims to deal with the case in which motif is the time series in a time series database which has the highest count of its similar time series within a range r. To adapt the above-mentioned method to the case in which motifs are frequently occurring subsequences of a longer time series, we modify MDTW_WedgeTree to a new algorithm for discovering “subsequence” motifs in time series under DTW. The proposed method consists of a segmentation method to divide the time series into motif candidates and a BIRCH-based clustering which can efficiently cluster motif candidate subsequences under DTW distance. Experimental results showed that our proposed method for discovering “subsequence” motifs performs very efficiently on large time series datasets while brings out high accuracy.

An Architectural Based Framework for the Distributed Collection, Analysis and Query from Inhomogeneous Time Series Data Sets and Wearables for Biofeedback Applications

The increasing professionalism of sports persons and desire of consumers to imitate this has led to an increased metrification of sport. This has been driven in no small part by the widespread availability of comparatively cheap assessment technologies and, more recently, wearable technologies. Historically, whilst these have produced large data sets, often only the most rudimentary analysis has taken place (Wisbey et al in: “Quantifying movement demands of AFL football using GPS tracking”). This paucity of analysis is due in no small part to the challenges of analysing large sets of data that are often from disparate data sources to glean useful key performance indicators, which has been a largely a labour intensive process. This paper presents a framework that can be cloud based for the gathering, storing and algorithmic interpretation of large and inhomogeneous time series data sets. The framework is architecture based and technology agnostic in the data sources it can gather, and presents a model for multi set analysis for inter- and intra- devices and individual subject matter. A sample implementation demonstrates the utility of the framework for sports performance data collected from distributed inertial sensors in the sport of swimming.

The Periodicity of Violent and Property Crime in Tshwane, South Africa

In this article, we identify and analyze the periodicity of violent and property crimes committed in Tshwane, South Africa, from 2001 to 2006. This is done using Fourier analysis, an advanced explorative mathematical technique commonly used in the physical sciences to detect the presence of a frequency or periodicity in a large time-series data set. The use of this technique in criminology is in its infancy, and in this study, Fourier analysis is used to identify periodic moments in time at which the risk of being a victim of violent and property crime in the city of Tshwane is heightened. Results indicated that violent crime peaks roughly every 7 and 75 days over the 5-year study period, with a marginal peak every 150 days. Property crimes peak every 75 days and every 150 days. Periodic peaks of crime observed in this study are explained using the central tenets of routine activities theory. Fourier analysis is an underused, powerful data-driven mathematical tool that should be added to the methodological arsenal available to criminologists when analyzing the temporal dimension of crime.

Interactive time series exploration powered by the marriage of similarity distances

Finding similar trends among time series data is critical for applications ranging from financial planning to policy making. The detection of these multifaceted relationships, especially time warped matching of time series of different lengths and alignments is prohibitively expensive to compute. To achieve real time responsiveness on large time series datasets, we propose a novel paradigm called Online Exploration of Time Series (ONEX) employing a powerful one-time preprocessing step that encodes critical similarity relationships to support subsequent rapid data exploration. Since the encoding of a huge number of pairwise similarity relationships for all variable lengths time series segments is not feasible, our work rests on the important insight that clustering with inexpensive point-to-point distances such as the Euclidean Distance can support subsequent time warped matching. Our ONEX framework overcomes the prohibitive computational costs associated with a more robust elastic distance namely the DTW by applying it over the surprisingly compact knowledge base instead of the raw data. Our comparative study reveals that ONEX is up to 19% more accurate and several times faster than the state-of-the-art. Beyond being a highly accurate and fast domain independent solution, ONEX offers a truly interactive exploration experience supporting novel time series operations.

Application of dynamic topic models to toxicogenomics data.

BackgroundAll biological processes are inherently dynamic. Biological systems evolve transiently or sustainably according to sequential time points after perturbation by environment insults, drugs and chemicals. Investigating the temporal behavior of molecular events has been an important subject to understand the underlying mechanisms governing the biological system in response to, such as, drug treatment. The intrinsic complexity of time series data requires appropriate computational algorithms for data interpretation. In this study, we propose, for the first time, the application of dynamic topic models (DTM) for analyzing time-series gene expression data.ResultsA large time-series toxicogenomics dataset was studied. It contains over 3144 microarrays of gene expression data corresponding to rat livers treated with 131 compounds (most are drugs) at two doses (control and high dose) in a repeated schedule containing four separate time points (4-, 8-, 15- and 29-day). We analyzed, with DTM, the topics (consisting of a set of genes) and their biological interpretations over these four time points. We identified hidden patterns embedded in this time-series gene expression profiles. From the topic distribution for compound-time condition, a number of drugs were successfully clustered by their shared mode-of-action such as PPARɑ agonists and COX inhibitors. The biological meaning underlying each topic was interpreted using diverse sources of information such as functional analysis of the pathways and therapeutic uses of the drugs. Additionally, we found that sample clusters produced by DTM are much more coherent in terms of functional categories when compared to traditional clustering algorithms.ConclusionsWe demonstrated that DTM, a text mining technique, can be a powerful computational approach for clustering time-series gene expression profiles with the probabilistic representation of their dynamic features along sequential time frames. The method offers an alternative way for uncovering hidden patterns embedded in time series gene expression profiles to gain enhanced understanding of dynamic behavior of gene regulation in the biological system.Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-016-1225-0) contains supplementary material, which is available to authorized users.

Mining precise cause and effect rules in large time series data of socio-economic indicators.

Discovery of cause–effect relationships, particularly in large databases of time-series is challenging because of continuous data of different characteristics and complex lagged relationships. In this paper, we have proposed a novel approach, to extract cause–effect relationships in large time series data set of socioeconomic indicators. The method enhances the scope of relationship discovery to cause–effect relationships by identifying multiple causal structures such as binary, transitive, many to one and cyclic. We use temporal association and temporal odds ratio to exclude noncausal association and to ensure the high reliability of discovered causal rules. We assess the method with both synthetic and real-world datasets. Our proposed method will help to build quantitative models to analyze socioeconomic processes by generating a precise cause–effect relationship between different economic indicators. The outcome shows that the proposed method can effectively discover existing causality structure in large time series databases.

What time is it? Deep learning approaches for circadian rhythms.

Motivation: Circadian rhythms date back to the origins of life, are found in virtually every species and every cell, and play fundamental roles in functions ranging from metabolism to cognition. Modern high-throughput technologies allow the measurement of concentrations of transcripts, metabolites and other species along the circadian cycle creating novel computational challenges and opportunities, including the problems of inferring whether a given species oscillate in circadian fashion or not, and inferring the time at which a set of measurements was taken.Results: We first curate several large synthetic and biological time series datasets containing labels for both periodic and aperiodic signals. We then use deep learning methods to develop and train BIO_CYCLE, a system to robustly estimate which signals are periodic in high-throughput circadian experiments, producing estimates of amplitudes, periods, phases, as well as several statistical significance measures. Using the curated data, BIO_CYCLE is compared to other approaches and shown to achieve state-of-the-art performance across multiple metrics. We then use deep learning methods to develop and train BIO_CLOCK to robustly estimate the time at which a particular single-time-point transcriptomic experiment was carried. In most cases, BIO_CLOCK can reliably predict time, within approximately 1 h, using the expression levels of only a small number of core clock genes. BIO_CLOCK is shown to work reasonably well across tissue types, and often with only small degradation across conditions. BIO_CLOCK is used to annotate most mouse experiments found in the GEO database with an inferred time stamp.Availability and Implementation: All data and software are publicly available on the CircadiOmics web portal: circadiomics.igb.uci.edu/.Contacts: fagostin@uci.edu or pfbaldi@uci.eduSupplementary information: Supplementary data are available at Bioinformatics online.

Process fault detection using time‐explicit Kiviat diagrams

Significant amounts of data are collected and stored during chemical process operations. The corresponding datasets are typically difficult to represent and analyze using traditional visualization methods. This article introduces time‐explicit Kiviat diagrams as a means to visualize the multidimensional time series data acquired from plant operations. This framework is then used to build multivariate control charts for large scale time series datasets, and to develop a fault detection mechanism that lends itself to real‐time implementation. The proposed methodology is applied to an industrial case study as well as to data obtained from the Tennessee Eastman process simulator, showing very good performance. © 2015 American Institute of Chemical Engineers AIChE J, 61: 4277–4293, 2015

Examining the spatial periodicity of crime in South Africa using Fourier analysis

This paper analyses the periodicity of violent and property crimes committed in Tshwane, South Africa from 2001 to 2006. Conventional monthly, daily and hourly variations in these types of crime are highlighted and then Fourier analysis is introduced as an advanced mathematical technique which is able to identify periodic peaks or signals of crime contained within a large time-series crime data-set. Specifically, Fourier analysis is used to examine crime periodicities in Tshwane in neighbourhoods stratified by social deprivation. Across most neighbourhood quintiles, there is a notable peak in violent crimes every 7–10 days. That is, every 7–10 days over the 5 years, there is a peak in violent crimes in most neighbourhoods in Tshwane. There is also a marginal peak occurring roughly every 50–60 days in neighbourhood Quintiles 1 (most affluent) and 3 (‘middle class’). Similar to violent crimes, the largest consistent periodicity for property crimes was observed across all five neighbourhood quintiles approximately every 7–10 days. A significant peak was also observed in property crime in Quintile 3 (‘middle class’) roughly every 75 days. Tentative explanations are provided for the results obtained.

A novel parallel scheme for fast similarity search in large time series

The similarity search is one of the fundamental components in time series data mining, e.g. clustering, classification, association rules mining. Many methods have been proposed to measure the similarity between time series, including Euclidean distance, Manhattan distance, and dynamic time warping (DTW). In contrast, DTW has been suggested to allow more robust similarity measure and be able to find the optimal alignment in time series. However, due to its quadratic time and space complexity, DTW is not suitable for large time series datasets. Many improving algorithms have been proposed for DTW search in large databases, such as approximate search or exact indexed search. Unlike the previous modified algorithm, this paper presents a novel parallel scheme for fast similarity search based on DTW, which is called MRDTW (MapRedcue-based DTW). The experimental results show that our approach not only retained the original accuracy as DTW, but also greatly improved the efficiency of similarity measure in large time series.

Multiple cell and population-level interactions with mouse embryonic stem cell heterogeneity.

Much of development and disease concerns the generation of gene expression differences between related cells sharing similar niches. However, most analyses of gene expression only assess population and time-averaged levels of steady-state transcription. The mechanisms driving differentiation are buried within snapshots of the average cell, lacking dynamic information and the diverse regulatory history experienced by individual cells. Here, we use a quantitative imaging platform with large time series data sets to determine the regulation of developmental gene expression by cell cycle, lineage, motility and environment. We apply this technology to the regulation of the pluripotency gene Nanog in mouse embryonic stem cells. Our data reveal the diversity of cell and population-level interactions with Nanog dynamics and heterogeneity, and how this regulation responds to triggers of pluripotency. Cell cycles are highly heterogeneous and cycle time increases with Nanog reporter expression, with longer, more variable cycle times as cells approach ground-state pluripotency. Nanog reporter expression is highly stable over multiple cell generations, with fluctuations within cycles confined by an attractor state. Modelling reveals an environmental component to expression stability, in addition to any cell-autonomous behaviour, and we identify interactions of cell density with both cycle behaviour and Nanog. Rex1 expression dynamics showed shared and distinct regulatory effects. Overall, our observations of multiple partially overlapping dynamic heterogeneities imply complex cell and environmental regulation of pluripotent cell behaviour, and suggest simple deterministic views of stem cell states are inappropriate.

Social contracts and community forestry: how can we design forest policies and tenure arrangements to generate local benefits?

There is widespread debate about the best strategies to provide local benefits in forest management. We evaluate recent policy changes in British Columbia, Canada, focussing on attempts to create local benefits from public forests through a community forestry program and broad policy changes in 2003 that removed obligations of tenure holders to process timber in areas near where timber was harvested. These obligations were intended to retain benefits of milling jobs locally and were considered part of a “social contract”. We evaluate these policy changes by asking two specific questions. (1) Do community forest tenures provide more local benefits than major industrial tenures? (2) How have the policy changes of 2003 affected the patterns of fibre flow over the period to 2008? We evaluate these questions through qualitative research and a quantitative fibre flow analysis using a large time-series dataset. Community forests as a group performed better than major industrial tenures in delivering local benefits as we defined them. However, large variation among individual community forests is evident, highlighting the disparate strategies used by communities to promote local benefits. Our fibre flow analysis did not reveal major changes following 2003, suggesting that broader fibre flow trends mask more local perturbations.

Clustering of large time series datasets

Time series clustering is a very effective approach in discovering valuable information in various systems such as finance, embedded bio-sensor and genome. However, focusing on the efficiency and scalability of these algorithms to deal with time series data has come at the expense of losing the usability and effectiveness of clustering. In this paper a new multi-step approach is proposed to improve the accuracy of clustering of time series data. In the first step, time series data are clustered approximately. Then, in the second step, the built clusters are split into sub-clusters. Finally, sub-clusters are merged in the third step. In contrast to existing approaches, this method can generate accurate clusters based on similarity in shape in very large time series datasets. The accuracy of the proposed method is evaluated using various published datasets in different domains.