|
|
Neural Network Based Edge Detection
Janda, Miloš ; Žák, Pavel (referee) ; Švub, Miroslav (advisor)
Aim of this thesis is description of neural network based edge detection methods that are substitute for classic methods of detection using edge operators. First chapters generally discussed the issues of image processing, edge detection and neural networks. The objective of the main part is to show process of generating synthetic images, extracting training datasets and discussing variants of suitable topologies of neural networks for purpose of edge detection. The last part of the thesis is dedicated to evaluating and measuring accuracy values of neural network.
|
|
|
Detection of Malicious Domain Names
Setinský, Jiří ; Perešíni, Martin (referee) ; Tisovčík, Peter (advisor)
The bachelor thesis deals with the detection of artificially generated domain names (DGA). The generated addresses serve as a means of communication between the attacker and the infected computer. By detection, we can detect and track infected computers on the network. The detection itself is preceded by the study of machine learning techniques, which will then be applied in the creation of the detector. To create the final classifier in the form of a decision tree, it was necessary to analyze the principle of DGA addresses. Based on their characteristics, the attributes were extracted, according to which the final classifier will be decided. After learning the classification model on the training set, the classifier was implemented in the target platform NEMEA as a detection module. After final optimizations and testing, we achieved a accuracy of the classifier of 99%, which is a very positive result. The NEMEA module is ready for real-world deployment to detect security incidents. In addition to the NEMEA module, another model was created to predict the accuracy of datasets with domain names. The model is trained based on the characteristics of the dataset and the accuracy of the DGA detector, whose behavior we want to predict.
|
|
|
Predikcia bankrotu lesníckych podnikov
Blihárová, Terézia
Blihárová, T. Bankruptcy prediction within the forestry sector. Bachelor thesis. Brno: Mendel University in Brno, 2023. This bachelor's thesis deals with bankruptcy prediction of companies in the for-estry sector in the EU using the binary logistic regression method. The aim of this thesis is to create bankruptcy models one, two and three years before the bank-ruptcy of companies of the positive group and subsequent verification of the cor-rectness of these models. A partial goal of this thesis is the optimization of the threshold in an effort to obtain the best possible classification capabilities of the models based on the classification of the bankrupt companies. The results section showed that the models with the optimal threshold have the ability to successfully predict the bankruptcy of companies in the forestry sector one and two years be-fore the upcoming bankruptcy. The model with the optimal threshold was able to predict the bankruptcy quite well even three years before the bankruptcy of the company, however, the rate of incorrect classification of active companies was relatively high.
|
|
|
Machine Learning-based Anomaly Detection in Industrial Control Systems
Tsymbal, Kateryna ; Holasová, Eva (referee) ; Pospíšil, Ondřej (advisor)
The main goal of this thesis is to design a system for anomaly and intrusion detection in industrial control systems using machine learning. The theoretical part of the thesis provides a basic theoretical overview of industrial control systems and their security. Furthermore, knowledge about anomaly detection techniques and potential challenges in this area are discussed. Lastly, the theoretical part has reviewed various solutions for anomaly detection in industrial control systems using machine learning. In the practical part, machine learning algorithms are applied to the selected HAI dataset. Finally, the findings on the suitability of the used algorithms and the possibilities for further research are summarized. The purpose of this thesis is to improve the security of industrial control systems, and the results can serve as a basis for the future development of more effective methods for anomaly detection in this area.
|
|
|
Detection of Malicious Domain Names
Setinský, Jiří ; Perešíni, Martin (referee) ; Tisovčík, Peter (advisor)
The bachelor thesis deals with the detection of artificially generated domain names (DGA). The generated addresses serve as a means of communication between the attacker and the infected computer. By detection, we can detect and track infected computers on the network. The detection itself is preceded by the study of machine learning techniques, which will then be applied in the creation of the detector. To create the final classifier in the form of a decision tree, it was necessary to analyze the principle of DGA addresses. Based on their characteristics, the attributes were extracted, according to which the final classifier will be decided. After learning the classification model on the training set, the classifier was implemented in the target platform NEMEA as a detection module. After final optimizations and testing, we achieved a accuracy of the classifier of 99%, which is a very positive result. The NEMEA module is ready for real-world deployment to detect security incidents. In addition to the NEMEA module, another model was created to predict the accuracy of datasets with domain names. The model is trained based on the characteristics of the dataset and the accuracy of the DGA detector, whose behavior we want to predict.
|
|
|
Identification Of Parkinson’S Disease Using Acousticanalysis Of Poem Recitation
Mucha, Ján
Parkinson’s disease (PD) is the second most frequent neurodegenerative disorder. It is estimated that 60–90% of PD patients suffer from speech disorder called hypokinetic dysarthria (HD). The goal of this work is to reveal influence of poem recitation on acoustic analysis of speech and propose concept of Parkinson’s disease identification based on this analysis. Classification methods used in this work are Random Forests and Support Vector Machine. The best achieved accuracy of disease identification is 70.66% with 59.25% sensitivity for Random Forests classifier fed mainly with articulation features. These results demonstrate a high potential of research in this area.
|
|
|
New Methodology Of Parkinsonic Dysgraphia
Analysis By Online Handwriting Using Fractional
Derivatives
Mucha, Ján
Parkinson’s disease (PD) is the second most frequent neurodegenerative disorder. One typical hallmark of PD is disruption in execution of practised skills such as handwriting. This paper introduces a new methodology of kinematic features calculation based on fractional derivatives applied on PD handwriting. Discrimination power of basic kinematic features (velocity, acceleration, jerk) was evaluated by classification analysis (using support vector machines and random forests). For this purpose, 37 PD patients and 38 healthy controls were enrolled. In comparison to results reported in other works, we proved that FDE in online handwriting analysis brings promising improvements. The best result of multivariate analysis was achieved with 83:89% classification accuracy in combination with 5 features using only one handwriting task (overlapped circles). This study reveals an impact of fractional derivatives based features in analysis of Parkinsonic dysgraphia.
|
|
|
Neural Network Based Edge Detection
Janda, Miloš ; Žák, Pavel (referee) ; Švub, Miroslav (advisor)
Aim of this thesis is description of neural network based edge detection methods that are substitute for classic methods of detection using edge operators. First chapters generally discussed the issues of image processing, edge detection and neural networks. The objective of the main part is to show process of generating synthetic images, extracting training datasets and discussing variants of suitable topologies of neural networks for purpose of edge detection. The last part of the thesis is dedicated to evaluating and measuring accuracy values of neural network.
|
guest ::
