Πλοήγηση ανά Επιβλέπων / ουσα "Alexopoulos, Angelos"

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 - 3 από 3

Forecasting in panel data
(2025-11-13) Pavelis, Konstantinos; Παβέλης, Κωνσταντίνος; Topaloglou, Nikolaos; Dendramis, Yiannis; Alexopoulos, Angelos
This study examines the use of econometric forecasting methods within a panel data framework, combining traditional approaches within modern dimensionality reduction techniques such as Factor Models, Fixed Effect Models and Principal Component Analysis (PCA). The research aims to demonstrate how the integration of information from panel data improves both the accuracy and explanatory power of forecasts for economic and financial variables, compared to classical time series methods. The empirical analysis is based on daily data from fifteen major global stock indices, covering the United States, Europe and Asia, for the period of 2014-2025. The dataset includes the variables Open, High, Low and Change, which are analyzed through econometric models such as Fixed Effect, Random Effects, and PCA Factor models. This analysis was conducted using the R programming language in the RStudio environment, taking advantage of its efficiency in processing and analyzing large and complex datasets. The results show that combining the panel data structure with dimensionality reduction techniques leads to significant improvements in predictive accuracy, offering deeper insights into the mechanisms driving market behavior. Indices such as S&P 500 and CAC40 displayed particularly high adjusted R2 values, indicating that a small set of latent factors extracted through PCA can effectively capture most of the variation in returns. Conversely, indices such as Nikkei 225 and Hang Seng showed lower explanatory power, suggesting that the presence of regional or idiosyncratic factors not fully captured by the models. The PCA Factor Models demonstrated that the first eight components were sufficient to explain approximately 80% of the variance in major indices, with PC3 and PC6 emerging as statistically significant predictors across most markets. Despite issues of heteroskedasticity and non normal residuals, the use of robust standard errors provided reliable inference, with no signs of multicollinearity or autocorrelation. Overall, the findings indicate that the combination of Fixed/Random Effects and Factor Models provides a comprehensive and effective framework for financial forecasting. This integrated approach yields statistically and theoretically sound results, enhances the understanding of global financial markets, and contributes to the development of evidence-based economic policymaking.
Machine learning framework for expected goals inference: bridging statistical modeling and causal understadning in football analytics
(2026-01-23) Troullinos, Michalis; Τρουλλινός, Μιχαήλ; Pagratis, Spyridon; Varthalitis, Petros; Alexopoulos, Angelos
Football analytics has evolved from descriptive statistics to sophisticated machine learning models, yet much of the existing research remains limited to isolated predictive or exploratory approaches. This study proposes an integrated analytical framework that combines supervised, unsupervised, and causal inference methodologies to advance the modeling of Expected Goals (xG) and deepen the understanding of shot quality and decision-making in football. Supervised models, including logistic regression, decision tree, random forests, gradient boosting, xgboost and neural network were employed to predict goal probabilities, while unsupervised methods such as Principal Component Analysis and K-Means clustering uncovered latent shot structures and tactical typologies. Isolation Forest identified rare or anomalous events, enhancing sensitivity to high-value but infrequent actions, and a Counterfactual Autoencoder introduced a causal layer that simulated “what-if” scenarios, quantifying how controlled changes in shot context, could affect the expected outcome. Experimental results demonstrate that ensemble models improve discriminative power, while unsupervised analyses reveal interpretable clusters that align with tactical behaviors on the pitch. The counterfactual component extends the framework beyond association, enabling causal insight into the determinants of chance creation. Collectively, these findings illustrate a progression from purely predictive modeling toward interpretable, context-aware, and prescriptive analytics. By bridging statistical accuracy with tactical relevance, this work contributes to the emerging paradigm of data-driven causal understanding in sports analytics, transforming raw performance data into actionable football intelligence.
A comparative study of autoencoders and GANs for credit card fraud detection
(2026-02-04) Karatzas, Konstantinos; Καρατζάς, Κωνσταντίνος; Pagratis, Spyridon; Dendramis, Yiannis; Alexopoulos, Angelos
Fraudulent behavior in the financial sector and its implications constitute a significant threat to the economy, society, and public trust. Detecting such behavior is therefore of crucial importance, as it enables timely intervention, and can potentially act as a deterrent. Despite this, fraud detection poses significant challenges, as it typically relies on limited labeled data and must operate under severe class imbalance. This thesis, using credit card transaction data, studies the fraud detection performance of autoencoders and GANs, while examining cross-model, architecture-level, and supervision-regime comparisons. By implementing a one-class, semi-supervised learning framework, where models are trained exclusively on non-fraudulent transactions, these deep learning models are employed to identify fraudulent transactions through various anomaly scoring and thresholding strategies, while exploring different optimization frameworks. Model performance is evaluated using the F1-score and confusion matrices, which are particularly appropriate for rare-event classification problems. The empirical analysis carried out in this study indicates that autoencoders exhibit stronger classification performance than GANs. It further shows that TPE-based optimization can yield better classification results when combined with the use of fraud labels during the optimization stage, compared to semi-supervised TPE optimization and manual search, highlighting the importance of hyperparameter selection. Overall, the findings emphasize the practical relevance of deep learning models, one-class training paradigms, and optimization method choice for fraud detection in highly imbalanced financial datasets.