Πλοήγηση ανά Επιβλέπων "Tsiamyrtzis, Panagiotis"

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω

Τώρα δείχνει 1 - 8 από 8

A Bayesian Kalman Filter approach in correcting near surface temperature forecasts
Raphael, Vasilia N.; Ραφαήλ, Βασιλεία Ν.; Athens University of Economics and Business, Department of Statistics; Tsiamyrtzis, Panagiotis
In Meteorology, Numerical Weather Prediction (NWP) models are used to provide forecasts for various weather related parameters. One such parameter, with high interest to the general public, is the near surface (2m) temperature. It is well known that the NWP based forecasts are inaccurate and various post processing methods exist, for improving the predictions. In this thesis a new method is proposed for the correction of near surface (2m) temperature forecasts provided by a NWP model. This procedure constitutes a Bayesian approach, based on a Kalman filter model. We provide an algorithm that results improved temperature predictions for future time-steps, by combining past observations and their corresponding NWP forecasts. The developed methodology is illustrated through the application of the proposed algorithm to a real data set, consisting of the observed and the forecasted temperatures of 700 days at a particular meteorological station (Thessaloniki).
Bayesian Statistical Process Control: Predictive Control Charts for continuous distributions in the regular exponential family
Bourazas, Konstantinos; Μπουραζάς, Κωνσταντίνος; Athens University of Economics and Business, Department of Statistics; Tsiamyrtzis, Panagiotis
In this thesis, the attention is focused in Statistical Process Control (SPC) with emphasis to phase I data. We will present a new general Bayesian self-starting procedure, which is based on the posterior predictive distribution and its name is Predictive Control Chart (PCC). We will analytically provide the initial assumptions and the construction of PCC. We will focus our attention on illustrations for single future observables of continuous distributions, which are members of the regular exponential family. A simulation study for out of control scenarios is used to evaluate and compare PCC against other sequential methods, either Frequentist or Bayesian, which are described analytically, for independent and normally distributed data. A sensitivity analysis for PCC concludes this thesis.
Bayesian Statistical Process Control: Predictive Control Charts for discrete distributions in the regular exponential family
Kiagias, Dimitriοs; Κιαγιάς, Δημήτριος; Athens University of Economics and Business, Department of Statistics; Tsiamyrtzis, Panagiotis
In this thesis a new general Bayesian self-starting method for Statistical Process Control (SPC) is presented. We will call this method Predictive Control Chart (PCC) and its construction will be based on the predictive distribution, as its name confess. This method can be used generally and especially when the likelihood is a member of the regular exponential family, where a conjugate prior will exist and the predictive distribution for a future observable can be obtained in closed form. We will focus our attention on illustrations for the discrete distributions of the regular exponential family in the univariate (parameter) case, while in multivariate cases the procedure will follow analogously. A simulation study is used to compare and evaluate the Predictive Control Chart (PCC) against Frequentist and Bayesian competing methods that exist in literature. A sensitivity analysis concludes this work.
CUSUM from a Bayesian perspective
Andreou, Maria Z.; Athens University of Economics and Business, Department of Statistics; Tsiamyrtzis, Panagiotis
The basic purpose of this thesis is to study the Classic Cumulative Sum method of Statistical Process Control (SPC), from a Bayesian prism. We consider the hypothesis testing problem of two possible models. Model under H0 will be the model of the in control situation and model under H1 will be the model of the out of control situation for a process. We wish to decide whether the model under H1, has been more plausible than the model under 𝐻0, or not. In other words, we want to decide whether a process has gone from an in control to an out of control situation. The CUSUM charting is designed to detect small but persistent shifts of a process’ characteristic as soon as possible, giving the optimal in control Average Run Length for a particular shift. From a Bayesian point of view, we consider the hypothesis testing problem as a decision theory problem, and using Bayes test theorem, we construct the Bayesian CUSUM. Due to the optimality properties of CUSUM (Moustakides (1986)), we aim to prove that the two forms are equivalent, thus, the Bayesian CUSUM we have constructed will be an optimal scheme for detecting a persistent shift in a process’ characteristic.
Modeling the return index in a production process of aluminum containers
(10/11/2018) Tsiagiannh, Christina I.; Τσιαγιάννη, Χριστίνα Ι.; Athens University of Economics and Business, Department of Statistics; Tsiamyrtzis, Panagiotis
In the present thesis we focus on performing statistical modeling in the production of refreshment containers (cans). The typical can is separated in 3 parts, which are called “Boby”, “End” and “Tab”. The three parts are produced separately and from a different aluminum alloy. This thesis is referred only to the production of the part “Body”. During the processing of aluminum a part of the material is cut off, which is next recycled in order to create new aluminum products. In various cases that some type of failure occurs, we need to scrap some of the material which goes for recycling and whose quantity the company wishes to minimize. One of the most important indexes of the plat is the Return Index which measures the internal recycling. Specifically, we take the ratio of the starting weight of aluminum over the weight produced in the end (original weight/ production). The main target is to find which variables affect significantly the return index. We start with a descriptive analysis and then move to multiple linear regression. It is known that the return index increases in the presence of non-conformities (failures). In this grounds, (Bayesian) logistic regression was applied to investigate which of the explanatory variables increase the probability of having a failure. Some of the explanatory variables were correlated and so decision trees were also tried. The statistical analysis performed using the freeware R.
Self-starting methods in Bayesian statistical process control and monitoring
(10/26/2021) Bourazas, Konstantinos; Μπουραζάς, Κωνσταντίνος; Ntzoufras, Ioannis; Demiris, Nikolaos; Psarakis, Stelios; Capizzi, Giovanna; Colosimo, Bianca Maria; Chakraborti, Subhabrata; Tsiamyrtzis, Panagiotis
In this dissertation, the center of attention is in the research area of Bayesian Statistical Process Control and Monitoring (SPC/M) with emphasis in developing self-starting methods for short horizon data. The aim is in detecting a process disorder as soon as it occurs, controlling the false alarm rate, and providing reliable posterior inference for the unknown parameters. Initially, we will present two general classes of methods for detecting parameter shifts for data that belong to the regular exponential family. The first, named Predictive Control Chart (PCC), focuses on transient shifts (outliers) and the second, named Predictive Ratio CUSUM (PRC), in persistent shifts. In addition, we present an online change point scheme available for both univariate or multivariate data, named Self-starting Shiryaev (3S). It is a generalization of the well-known Shiryaev’s procedure, which will utilize the cumulative posterior probability that a change point has been occurred. An extensive simulation study along with a sensitivity analysis evaluate the performance of the proposed methods and compare them against standard alternatives. Technical details, algorithms and general guidelines for all methods are provided to assist in their implementation, while applications to real data illustrate them in practice.
Sentiment analysis using statistical methods
Vryniotis, Vasileios; Βρυνιώτης, Βασίλειος; Athens University of Economics and Business, Department of Statistics; Tsiamyrtzis, Panagiotis
Sentiment analysis is the task of identifying automatically the polarity, the subjectivity and the emotional states of particular document or sentence. Sentiment analysis became increasingly important after the rise of social media networks and it has applications in various fields including marketing, crisis management, search engines and semantic web. In this thesis, we will use various text classification techniques to detect the language of a given document and to evaluate its polarity. We will examine in detail several algorithms such as Naïve Bayes, Max Entropy and Multinomial Logistic Regression. We will argue regarding the importance of the introduction of the neutral class and we will evaluate their performance on commonly used datasets.
Statistical process control and monitoring with big data
(10/12/2018) Kokkinopoulou, Xeni D.; Κοκκινοπούλου, Ξένη; Athens University of Economics and Business, Department of Statistics; Tsiamyrtzis, Panagiotis
Στην παρούσα διατριβή, εξετάζουμε πώς συμπεριφέρονται τα κλασσικά διαγράμματα ελέγχου με μεγάλα δεδομένα, τόσο όταν μια διαδικασία είναι εντός ελέγχου όσο και όταν αυτή είναι εκτός ελέγχου. Επιπρόσθετα εξετάζονται και κάποια εναλλακτικά σχήματα ελέγχου. Κάποια από αυτά περιλαμβάνουν τον έλεγχο Kolmogorov - Smirnov, ένα μη παραμετρικό έλεγχο λόγων πιθανοφάνειας για στοχαστικά διατεταγμένες τυχαίες μεταβλητές καθώς και την χρήση των διαγραμμάτων ποσοστιαίων σημείων (Q-Q plots). Όλες οι προαναφερθείσες μεθοδολογίες είναι μη παραμετρικές, προκειμένου να ωφεληθούμε όσο το δυνατόν περισσότερο από τους μεγάλους όγκους δεδομένων. Καταλήγουμε ότι τα Q-Q plots είναι η πιο αποτελεσματική μεθοδολογία σε καταστάσεις εντός αλλά και εκτός ελέγχου μιας διαδικασίας.