|18 - 19 March 2022, Ishikawa, Japan|
|Final extension - Firm submission deadline: 22 November 2021|
We understand a long time-series through features and trends and their transitions, for example, “Globally it increases a little, but it starts with a medium value, decreases a little in the beginning and has big oscillations at end.” It is often the case where the periods of features and trends are determined by the data themselves. We must, therefore, partition time-series into several periods of different features and trends.
We propose a method to partition time-series data by clustering adjacent data with the total similarity of their values, changes of values and degrees of oscillations of adjacent periods. First we have the initial clusters of line segments of adjacent data in time. Next we get the adjacent clusters that have the maximum total similarity and merge them into one. We repeat this process until the condition of termination. We formulate the total similarity as the weighted average of three similarities of the value, change of values and degree of oscillations. The weights are very important. The fixed weights cannot have the clustering results that fit our sense. We, therefore, propose variable weights with three similarities and sizes of adjacent clusters with the operation of ordered weighted average. Furthermore, in order to exclude small clusters of outliers, we define similarities of two clusters adjacent to the small cluster. We apply this method to actual time-series data and show results. The method can improve linguistic expressions of time-series data and retrieval of similar time-series with linguistic similarity.
Motohide Umano received the B.S. degree in 1974, the M.S. degree in 1976 and the Dr. of Engineering degree in 1979 in Information and Computer Sciences from Osaka University, Japan. He was an Assistant Professor of Department of Applied Mathematics, Faculty of Science, Okayama University of Science in 1979-1985. Then he moved to Osaka University, where he was a Research Associate, an Assistant Professor and an Associate Professor of Computation Center in 1985-1990, an Associate Professor of Department of Precision Engineering, Faculty of Engineering in 1991-1993 and Department of Systems Engineering, Faculty of Engineering Science in 1993–1996. He then moved as a Professor to Osaka Prefecture University, where he was a member of Department of Mathematics and Information Sciences, College of Integrated Arts and Sciences in 1996-2005 and Department of Mathematics and Information Sciences, Graduate School of Science in 2005-2016 and Department of Computer Science and Intelligent Systems, Graduate School of Engineering in 2016-2017 and he took Professor Emeritus of Osaka Prefecture University. He is a Technical Adviser of Intelligent Machinery Research Center, Technical Research Institute, Hitachi Zosen Corporation.
His current research interests are fuzzy data/knowledge information processing including fuzzy-set manipulation systems, fuzzy database systems, fuzzy intelligent systems and learning of fuzzy knowledge from data.
Multiple Criteria Decision Aiding (MCDA) is constituted by a set of concepts, techniques and procedure aiming to provide a recommendation in complex decision contexts. MCDA is based on a constructive approach that aims to build a preference model in cooperation between the analyst and the Decision Maker. A typical MCDA methodology is the ordinal regression aiming to define a decision model in a given class (an additive value function, a Choquet integral, an outranking model such as ELECTRE or PROMETHEE and so on) representing the preference information provided by the DM. Recently ordinal regression has been extended and generalized through Robust Ordinal Regression taking into account the idea that there is a plurality of decision models in a given class compatible with the preferences expressed by the decision maker. Originally, the set of compatible decision models was used to define the necessary and possible preference relations holding when the preference holds for all value functions or for at least one value function, respectively. After, a probability distribution on the set of compatible decision model was introduced to define probabilistic preferences. ROR has been also fruitfully applied to interactive optimization procedures. In this talk I shall present the basic concepts, the principal models, the main applications and the recent developments of Robust Ordinal Regression taking into consideration its advantages in the context of an MCDA constructive approach.
Salvatore Greco is full professor at the Department of Economics and Business at the University of Catania where has been teaching Decision Theory, General Mathematics, Financial Mathematics and Actuarial Mathematics. Since 2013 Salvatore Greco has also a part time position at the Business School of Portsmouth University (UK). His research regards preference modeling and multiple criteria decision analysis (MCDA) with a specific attention to application of rough set theory, non-additive integrals, evolutionary multiobjective optimization methodologies, composite indices for sustainable development, wellbeing and innovation, MCDA models for territorial and urban planning.
At the 22nd International conference on MCDM held in Malaga June 17-22 2013, he received the MCDM Gold Medal being “the highest honor that the International Society on Multiple Criteria Decision Making bestows upon a scholar who, over a distinguished career, has devoted much of his/her talent, time, and energy to advancing the field of MCDM, and who has markedly contributed to the theory, methodology, and practice of MCDM”
Since 2010, Salvatore Greco is one of the three coordinators of the EURO Working Group in Multiple Criteria Decision Aiding. He has been member of the executive committee of International Society on Multiple Criteria Decision Making (http://www.mcdmsociety.org/) for the years 2006-2009, 2011-2013, 2016-2019. In the years 2014-2019 Salvatore Greco was member of the scientific committee of AMASES (Italian Society for mathematics applied to economics and social sciences and in the years 2017-2019 he served as vicepresident. He is currently the president elect of the MCDM section of INFORMS.
Scopus reports 248 publications of Salvatore Greco cited all together 9734 times and an h-index of 51. Google Scholar reports a total of 2223 citations with an h-index of 68.
Many traditional optimization models are limited with respect to the management of inconsistency that frequently appear between decision and optimization goals. As consequence, such models in many cases achieve results that may be optimal for the model but are not for the use case to be managed. In real world use cases both decision and optimization goals are usually partly conflicting and therefore partly inconsistent. Assumptions like independence of goals, additivity or monotonicity as preconditions usually do not hold. Due to this, traditional concepts like integration based on weighted sums, for instance, in many cases do not really help. In this talk we describe some applications of a decision and optimization model based on (extended fuzzy) interactions between goals (DMIG) to some relevant real-world decision and optimization use cases. After a brief discussion of the basics of the concept of the model, example use cases are presented and advantages of their solutions are shown. The use cases are related to real-world decision and optimization problems in business processes such as management and scheduling of field forces that maintain complex industrial infrastructure, management of resources based on sequencing of production orders in car producing factories and automated management of bus and tram depots. Some additional examples are named. It is also shown how the so-called key performance indicators (KPIs) of such real-world use cases are understood as decision and optimization goals and how interactions between decision and optimization goals build a bridge to the optimization of real-world KPIs. Finally, it is discussed why DMIG may be used for learning of consistent preferences between the KPIs and how the concept is connected to the field of machine learning.
After studying computer science and business administration at the University of Dortmund, Dr. Rudolf Felix did his doctorate in the field of decision support systems and fuzzy logic. In 1992 he founded PSI FLS Fuzzy LogiK & Neuro Systeme GmbH and continuously developed PSI FLS' own Deep Qualicision KI technology, which is based on machine learning and neural networks as well as on qualitative labeling using optimization algorithms.
The solutions are successful in sectors such as the automotive industry, the automotive supplier industry, the energy industry, transport logistics, local public transport or retail as a cross-sectional technology in productive use. Qualicision technology has been complementing the software tools and applications of the Berlin PSI Group since 2008. Dr. Rudolf Felix published more than 60 scientific and numerous user-oriented articles in various specialist journals. In addition to his managerial activities in the integration of Qualicision into complex business processes in a multi-criteria environment, he is a member of the European Society of Fuzzy Logic and Technology, EUSFLAT.