Dissertation Title: PREDICTION ANALYSIS AND CLASSIFICATIONS IN MACHINE LEARNING USING BAYESIAN THEORY: MULTIDIMENSIONAL MIXED BETA AND OTHER MODELS IN MICRO-EXPRESSION RECOGNITION
- Authors: Viola BAKIASI
- Institution: University of Tirana, Faculty of Natural Sciences, Department: Applied Mathematics
- Field of study: Applied Mathematics
- Publication date: 03/06/2026
- The dissertation is published in Albanian.
© Copyright: Viola BAKIASI
Published by the University of Tirana. Based on legal acts, regulations and policies of the UT.
👉Click here to view the full dissertation (PDF)
Briefing
Automatic recognition of facial micro-expressions poses a significant challenge in affective computing due to the very low visual intensity, short duration of the emotional signal, and class imbalance in existing datasets. This dissertation addresses this problem by developing a hybrid methodological framework that integrates dimensionality reduction, supervised classification, and Bayesian inference into a unified process.
The methodology includes image preprocessing, feature extraction, and dimensionality reduction using Principal Component Analysis (PCA), Kernel Principal Component Analysis (KPCA), and t-Nearest Neighbor Stochastic Distribution (t-SNE). Various supervised learning classifiers are applied to these reduced representations, including Support Vector Classifier (SVM), K-Nearest Neighbors (K-NN), Random Forest (RF), Naive Bayes, and combined classification. Bayesian inference is used to calibrate classification probabilities and quantify prediction uncertainty.
The system is implemented in Python and tested on the AffectNet and CASME II datasets, and a Graphical User Interface (GUI) for real-time microexpression detection has been developed. The results show that combining dimensionality reduction with supervised classifiers and probabilistic modeling improves the stability of performance and the balance between accuracy and F1 value under conditions with unbalanced data.
Keywords: micro-expression, affective computing, dimensionality reduction, supervised learning, combinatorial classification, Bayesian inference, GUI.
BEXTRACT
Automatic recognition of facial micro-expressions represents a significant challenge in affective computing due to their extremely low visual intensity, short duration of the emotional signal, and class imbalance in existing datasets. This dissertation addresses this problem through the development of a hybrid methodological framework that integrates dimensionality reduction, supervised classification, and Bayesian inference within a unified pipeline.
The methodology includes image preprocessing, extraction of discriminative features, and dimensionality reduction through Principal Component Analysis (PCA), Kernel Principal Component Analysis (KPCA), and t-Distributed Stochastic Neighbor Embedding (t-SNE). On these reduced representations, several supervised learning classifiers are applied, including Support Vector Machines (SVM), K-Nearest Neighbors (K-NN), Random Forest (RF), Naive Bayes, and stacking. Bayesian inference is employed for probability calibration and quantitative uncertainty estimation of classification predictions.
The system is implemented in Python and evaluated on the AffectNet and CASME II datasets. Additionally, a Graphical User Interface (GUI) has been developed for real-time micro-expression detection. The results demonstrate that combining dimensionality reduction with supervised classifiers and Bayesian calibration improves performance stability and enhances the balance between accuracy and F1-score under class-imbalanced conditions.
Keywords: micro expressions, affective computing, dimensionality reduction, supervised learning, stacking, Bayesian inference, GUI.