Dissertation Title: COMPARISON OF HYPERPARAMETERS WITH OPTIMIZATION METHODS USING ANALYSIS BEJESIANE 

Authors: Blerina BOÇI 

• Institution: University of Tirana, Faculty of Natural Sciences, Department: Industrial Chemistry

• Field of study: Natural Sciences, Applied Mathematics

• Publication date: 10/06/2026

• The dissertation is published in Albanian.

© Copyright: Blerina BOÇI 

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
Bayesian optimization is an advanced technique that has found use in solving
of global optimization problems, with applications in various fields, such as science,
engineering, economics and beyond. This method is particularly useful in cases
when the objective function is costly to estimate and is not known in the form
analytical. Development of computer systems and increased use of intelligence
artificial intelligence have increased the need for the use of various optimization methods. In
machine learning, hyperparameter optimization is an essential process for
improving the performance of models. In this regard, Bayesian optimization
is considered an effective method, which plays an important role in learning
automatic, enabling automatic optimization of hyperparameters. This method
is based on probabilistic models, usually Gaussian processes, which are used to
approximate the objective function and to assess the uncertainty of the predictions.
Through acquisition functions, BO explores new areas of the search space.
and exploits previously identified promising areas. This process is based on
information obtained from previous evaluations of the objective function, in order to
finding the optimal solution.
This dissertation focuses on the optimization of hyperparameters of learning models.
automatic and in optimizing the parameters of optimization algorithms.
The proposed methodology is applied in three main directions: (i) in regression models for
forecasting salaries with real data from INSTAT; (ii) in classification models by
using demographic data and data from a student survey; (iii) in
optimization of parameters of evolutionary algorithms DE and PSO in problems
engineering with limitations. Experimental results show significant improvements
in performance and durability, confirmed through tests and analyses
statistical. These results confirm the effectiveness of Bayesian optimization in
dealing with complex problems in real contexts.
Keywords: Bayesian optimization, machine learning, regression, classification, model,
parameter, hyperparameter