In this age of big data and the availability of many speedy stylized algorithms including deep learning algorithms, there has been a tremendous increase in the number of manuscripts on time series clustering and classification in such diverse fields as economy, finance, environment science, computer science, engineering, physics, seismology, hydrometeorology, robotics, biology, genetics, neurology and medicine.
This topic can be categorized into two parts.
The first part is unsupervised and commonly referred to as clustering. No knowledge of prior groupings is available. For instance, in finance, it is interesting to cluster a set of companies quoted in the Market Exchange by analyzing the performances of the shares in a time period; in this way, the results of the cluster analysis can represent useful information for planning, at the right time, the policy and the strategic assessment of the establishment and the behaviour of the operators that invest money in stocks. In economics, it is very important to cluster different countries by considering the dynamics of the economic aggregates. In marketing research, it is interesting, to support companies’ decision strategies, to assess analogies or diversities among flavours, preferences and consumption behaviour of consumers at different times. In anthropometry, psychometrics, sociology and behavior studies, it is interesting to cluster individuals based on their moods, emotions and behavior when observed at different times.
In this framework, time series clustering can be grouped in three theoretical classes:
Observation-based (or raw data-based) clustering: it relies on raw data to conduct the cluster analysis, by using suitable metrics based on cross sectional and/or longitudinal characteristics.
Model-based clustering: it considers the features of the models fitted to the time series, e.g. ARIMA models, GARCH models, TAR models, splines, distribution models, functional models.
Feature-based clustering: it relies on features derived from the observed time series, e.g. autocorrelation, quantile autocovariance, cross correlation, periodogram and its transformations, coherence, wavelets, cepstral.
The second part is supervised and commonly referred to as classification, in which case knowledge of prior groupings is available. Classification is a supervised approach to grouping together items of interest and discriminant analysis, neural networks and machine learning methods are amongst the methodological approaches that are used. Time series classification methods include predominantly the use of feature-based, model-based and machine learning techniques. The features are extracted in the time domain, the frequency domain and the wavelet domain. Model-based approaches for time series classification include ARIMA models, Gaussian mixture models and Bayesian approaches, while machine learning approaches include classification trees, nearest neighbour methods and support vector machines.
Example of time series classification is the discriminant analysis of electrocardiogram (ECG) signals for three leads based on a three-dimensional formulation of a single dipole of the heart.
Contributions on both topics are welcomed from academic researchers in any field, analysts in statistics, data science, computer science, engineering, economics and finance, business and industry, management and marketing, medicine, environment science and hydrology, physics, biology and genetics, and many others.
Topics of particular interest include, but are not limited to:
· Hierarchical and non-hierarchical clustering of time series
· Observation-based clustering of time series
· Feature-based clustering of time series
· Model-based clustering of time series
· Fuzzy clustering for time series
· Robust clustering for time series
· Mixture clustering models for time series
· Self-Organizing Maps for time series
· Hidden Markov models for time seriess
· Support Vector machines for time series
· Classification trees for time series
· Discriminant analysis for time series
· Bayesian clustering and classification for time series
· Neural networks for time series
· Nearest neighbours methods for time series
· Unsupervised learning for time series
· Supervised learining for time series
· Clustering and classification of big time series
· Clustering and classification of interval time series
Submission details: The full paper should not exceed approximately 20 pages (A4 or Letter size with 12 point), including illustrations and tables. The front page of the manuscript must contain: a concise and informative title; the names, affiliations, postal and e-mail addresses of all authors; telephone and fax number of the corresponding author; an abstract of 8–10 lines; and 4–6 keywords which can be used for indexing purposes. Further formatting instructions are given on the journal’s homepage (Guide for Authors):
The manuscript should be submitted electronically as a pdf file via Elsevier’s Editorial Manager https://www.editorialmanager.com/IJA/default.aspx and needs to be specified as a paper for this special issue.
The review process will be similar to that for regular issues of the journal.
• The date the first paper is expected (submission portal will be open from this date): September 1, 2020.
• The submission deadline (the date by which all papers should be submitted to the Guest Editors for review and the EM submission site will be closed): December 1, 2020.
• The acceptance deadline (the date by which all manuscripts should be fully reviewed and final decisions made on all manuscripts): September 1, 2021.
Pierpaolo D’Urso, Sapienza – University of Rome, Italy
Livia De Giovanni, Luiss Guido Carli University, Rome, Italy
Elizabeth Ann Maharaj, Monash University, Melbourne, Australia