Σχεδιασμός, υλοποίηση και δοκιμή τεχνικών μηχανικής μάθησης για την ανίχνευση κυβερνοεπιθέσεων

Σταυρόπουλος, Γεώργιος

Design, implementation and testing of machine learning techniques for cyberattack detection

Master Thesis

Author

Σταυρόπουλος, Γεώργιος

Date

2025-04

Abstract

Cyberattack detection using Machine Learning has emerged as one of the most dynamic fields of research, enabling the real-time analysis of large datasets and the identification of abnormal be haviors with significantly greater accuracy compared to traditional methods. This study focuses on the classification of malware represented images, into five distinct categories: Backdoors & RATs, Downloaders & Droppers, Multipurpose, Spyware & Adware, Rogue Software & Fraud ware, and Worms & Self-Replicating Malware. A Transfer Learning approach was adopted, utiliz ing features extracted from the publicly available Malimg dataset through the integration of two Pre-Trained on ImageNet deep learning architectures — ResNet-50 and DenseNet-121. Bayes ian Optimization was employed to determine the optimal number of layers to unfreeze for fine tuning, enhancing the quality of feature extraction. Subsequently, feature selection was performed using the ANOVA method. For classification, three model families were explored: a non-linear model (Support Vector Classification – SVC), a neural network (Multi-Layer Perceptron – MLP), and a tree-based ensemble model (CatBoost). All models underwent hyperparameter optimiza tion via Bayesian search. Results demonstrated exceptional performance, with all evaluation met rics — including per-class (accuracy, precision, recall, F1-score, specificity, ROC AUC), averaged (macro, weighted), and global (micro metrics, balanced accuracy, Cohen’s Kappa, Matthews Cor relation Coefficient) consistently exceeding 98%, and in most cases, 99%. Finally, model combi nation using the Borda count method further improved accuracy, yielding only 35 misclassifica tions out of a total of 8,024 samples.

Postgraduate Studies Programme

Κυβερνοασφάλεια και Επιστήμη Δεδομένων

Department

Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Πληροφορικής

Number of pages

421

Language

Greek

URI

https://dione.lib.unipi.gr/xmlui/handle/unipi/17703

Collections

Τμήμα Πληροφορικής

Show full item record

Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα