Theses supervised by Yrd. Doç. Dr. Abdullahı Abdu Ibrahım
17 theses · Altınbaş University
Akıllı evlerde ıot tabanlı akıllı ölçüm sistemi
Even while new technologies are being developed to aid in the growth of humanity, their security systems still have weaknesses that might be exploited by those who want to steal or damage the data of others. An example of a typical security issue is the usage of passwords for devices that are either too easy to crack or too simple for users to remember. Botnets are a tactic used by malicious actors; they are networks of compromised devices that may be used to execute commands. Botnets are networks of hacked computing devices. As a direct result, issues about the security of the Internet of Things (IoT) must be addressed. According to, there might be as many as 26 billion Internet of Things (IoT) devices linked to the internet by the year 2020. Therefore, it is crucial to develop solutions for safeguarding the Internet of Things' security (IoT). In many network settings, intrusion detection systems serve as the first line of defense for network security. Several studies have been undertaken in this field of study, including [8] and [9], which use supervised machine learning to identify instances of hazardous behavior. [8] and [9] In these sorts of systems, the process of tagging databases demands a large investment of both computational and human resources. During the dataset preparation phase, also known as step 1, this technique's algorithm picks features based on the qualities of the data sets in order to organize them in decreasing order of their degree of similarity. This procedure is known as feature selection. DoS, R2L, U2R, and probing are among the four types of attacks that are included in the KDD CUP 99(KDD) dataset, which was created for testing IDS. There are 41 KDD features that are unique to every TCP connection. These characteristics may be categorized into three groups: traffic features, fundamental features, and content features. KDD has been used often in situations requiring intrusion detection data mining and machine learning investigations.
Mikro şerit tasarımı ve geliştirilmesi kişisel ağlar için yama antenidevlet kurumlarında
In our daily lives, especially in the past few years, wireless communications have become an important and effective era in performing tasks and communicating wirelessly without the need for a physical medium. The most important feature of remote communications is the ease of use and installation in places where it is difficult to communicate physical media, and the possibility of sending information over large distances by employing electromagnetic waves. With this rising in the number for the appliances and the implementations connected to the wireless network, there has been crowding, which negatively affected the speed of data transmission. One of the effective solutions proposed by network operators is to switch to the use of high frequencies to meet the users' desire for high data transmission speed. The antenna is considered as a one of the considerable essential elements in the wireless communication systems. The microstrip antennas are one of the types of antennas used, but they suffer from weak gain and narrow bandwidth, so there is an importance to improve these parameters. In this paper, a teeny sized MPA is simulated and optimised to be operated at the 62 GHz frequency for the personal network applications within the government institutions based on the CST antennas modelling software. In order to improving the MPA performance, two shapes of the Frequency Selective Surface (FSS) are used. The FSS is acting as a passive filter to make the antenna more intelligent because it is selectively passing the preferred frequencies or rejecting the unwanted ones.
Merkezi olmayan güvenlik ve verilerin kullanım blokzincirinin bütünlüğü öğrenme teknikleri
Blockchain technology has been linked with Deep Learning for a long time now. There are many issues that are hinder the implementation of deep learning applications at a large scale. Surveys and studies from multiple sources reveal that security threats and data privacy are still the primary concerns. These problems are well known and solutions exist for these problems in the IT industry. However, traditional IT security solutions cannot be applied to Deep learning for various reasons spanning from type of devices to sheer volume of devices. Unfortunately, like in any other industry, security is often disregarded in the deep learning domain as well, and most of the resources are allocated to application development and device hardware. So, the search for a silver bullet to overcome these inhibitors has been going on for a while. After Bitcoin became prominent, people started to realize the potential of the underlying distributed ledger (blockchain) technology and considered it as a true innovation. Rather than facilitating a peer-to-peer digital payment system involving a cryptocurrency, the blockchain technology is viewed as a mechanism that provides device identity, secure data transfer, and immutable data storage. All these features can be implemented without any centralized authority and a completely transparent system with auditable cryptographic proofs. Our aim through this research thesis is to get a deep level ABSTRACT DECENTRALIZED SECURITY AND DATA INTEGRITY OF BLOCKCHAIN USING DEEP LEARNING TECHNIQUES AL-KHAFAJI, Ali Khaleel Ibrahim, M.Sc., Electrical and Computer Engineering, Altınbaş University, Supervisor. Asst. Prof. Dr. Abdullahi Abdu IBRAHIM Date: 8 / 2022 Pages: 65 viii understanding of the blockchain technology and study some of the widely used blockchain frameworks including Ethereum, Bitcoin, and Litecoin. We will further examine the exclusive features offered by each of these frameworks and define their target use cases. While researching about each framework, we plan to deploy a blockchain in the local network i.e., private blockchain and operate on it from different devices running on various operating systems. In each deployment, we will observe the functional issues and benchmark system requirements for running different types of nodes. Also, we will study different algorithms involved in each framework, compare them with each other, and derive their suitability for Deep learning. Ultimately, our aim is to determine the most suitable blockchain architecture for the Deep learning ecosystem. A high-level comparison of the researched architectures will be provided so that managers and developers can quickly decide on a suitable framework for their application or use case depending upon the requirements. For each architecture, a set of sample use cases and on-going research will be discussed to get an idea of the usage of that architecture in the real world.
Kendi için mikro şerit anten tasarımı ve üretimi güç modern kablosuz sistem
Due to the urgent need for modern antenna technology to fit the recent applications including 5G systems, IoT terminals, and space satellite communication systems. However, the antenna technology still rarely relaxed due to the antenna size and passive performance limitations. This involves many advances to include self-powered systems based on green energy based on solar and radio frequency resources. Therefore, novel-reconfigurable sub-6 GHz microstrip patch- antenna operating on three resonant frequencies 3.6, 3.9, and 4.9 GHz is designed for 5G- applications. The proposed antenna is structured from metamaterial-(MTM) array with matching circuit printed around a printed strip line. Antenna is excited with a coplanar waveguide to achieve an excellent-matching over a wide frequency band. The suggested antenna shows-excellent-performance in expression of S11, gain, and radiation-pattern that are controlled by two photo-resistances. Suggested antenna shows various operating-frequencies and radiation-patterns after changing the photo-resistance cases. The main antenna novelty is realized by splitting the main-lobe which tracks more users on same operating-frequency. Nevertheless, major radiation-lobe can be steered to required-location by controlling the surface current motion utilizing two varactor-diodes on the matching circuit. Next, the proposed design is developed to 3D array geometry for 5G-applications. The suggested antenna-array provides wideband with excellent matching-impedance, S11≤-10dB, from 3.1GHz to 5.75GHz. The suggested MIMO array is constructed from four elements arranged on cubical structure to provide low mutual-coupling, less than -20dB, over all interested frequency bands. Each element is excited with coplanar-waveguide-(CPW). The radiation patterns are controlled by two optical-switches of Light Dependent Resistors-(LDRs). The suggested array shows excellent response to LDR statue changes in terms of directivity. The proposed antenna frequency band is found to be insignificantly affected with respect to LDR switching statues. The suggested array is fabricated and tested experimentally on expressions of S-parameters, gain and radiation-patterns at 3.6GHz, 3.9GHz, and 4.9GHz. The maximum gain is found to be 3.6dBi, 4.2 dBi, and 4.1dBi at 3.6GHz, 3.9GHz, and 4.9GHz, respectively. Nevertheless, it is found that the suggested array achieved a significant-beam steering about ±30o, ±40o, and ±45o at 3.6GHz, 3.9GHz, and 4.9GHz, respectively, after changing LDR switching statues. Such technique ensures the antenna beam controlling without need for biasing circuits. Next, RF energy harvesting system is proposed for self-powered system with solar panel integration. At the end, the antenna array is designed and fabricated, the simulation is done by using CST Studio software. It is found that the experimental results of fabrication are agreed very well with those obtained from simulations.
Güç tüketimi tahmini için yapay sinir ağları algoritmalarının matlab uygulaması
At the same time as there is a drive to increase electrification efforts on a global scale, new technological innovations are leading to an increase in the amount of power that is consumed worldwide. As a result, we should anticipate an increase in the amount of electricity used inside the residential setting. It is becoming increasingly difficult to generate reliable projections about future power consumption as more information about the usage of energy throughout the world becomes available. Both the buyer and the seller stand to benefit from an accurate prognosis. A customer may be able to save money and reduce their carbon footprint with the assistance of a power forecast. It is essential for the effectiveness of the provider's attempts to regulate the flow of goods that they have a forecast that can be relied upon. Consequently, this type of modeling can make a contribution to the overall optimization of the supply chain in the residential electricity industry. As a result of the extensive interest in this issue, a broad variety of techniques to solving it have been investigated and assessed. In this study, we introduce the AAA-ANN and ANN-PSO algorithms for accurately predicting energy consumption. Both of these methods are based on artificial neural networks. In order to provide accurate projections of future power consumption, the algorithms underlying both approaches were developed in Matlab and then trained using data taken from the UCP database. The AAA-ANN method outperforms other approaches in terms of error curve analysis, training accuracy rate analysis, vitesting accuracy rate analysis, training error rate analysis, and testing error rate analysis. Afterc comparing AAA-ANN to ANN and ANN-PSO, it became abundantly evident that this particular method was the best approach for forecasting future power use.
Lowpan kullanarak kablosuz sensör ağı için enerji farkındalığını artırın
There are many networks in remote areas to support all areas and applications. Networks That allow connectivity covering a range of square kilometers are critical for these remote deployments. The widely used star topology is not ideal for a rural environment as the coverage is limited by the central axis positioning which also contributes to being a single point of failure. It is clear that mesh networks are more attractive in this respect, but scalability has always been an issue for mesh networks, especially with regard to routing. Saving power very fundamental in remote IoT deployments, where devices can be left in isolated fields for an extended period of time. In this thesis we dealt with the steering problem of remote sensors by introducing reinforcement learning energy-aware routing algorithms. We determined the strength of the RL routing algorithm for remote sensing networks. This thesis also presents a step-by-step detailed analysis of the RL routing algorithm to To prove the effectiveness of the algorithm. We model the network operating in a region bounded by a finite number of randomly distributed nodes within the region in this algorithm. Hence, we have defined the service area of the target network assuming network limitations in the model.
Makine öğrenmeyi kullanarak sosyal medyadaki kötü amaçlı URL'leri tespit ve sınıflandırma
Recently, the variety and size of malware on the social networks has increased dramatically, bearing phrases and headings aimed at attracting attention and pushing them to enter the link that contains malicious software, causing theft (bank accounts, financial transactions, installing malicious software) and therefore it is necessary to discover These risks and threats are addressed. The purpose of this thesis is to discover malware and classify it into benign or malicious URLs using a machine learning algorithm called Support vector machine, which is used in binary classification and the algorithm was utilized for creating a model for malware detection. This study is conducting comprehensive experiments for the purpose of comparing and verifying the suggested method's results with the ones of other techniques. The experimental results are showing that the presented approach is achieving strong detection and high accuracy of up to 93%, and it is a method that achieves strong detection compared to other results for detecting malware.
An improved agent protection model for distributed denial of service flooding attack traffics
Dağıtılmış Hizmet Reddi (DDoS), interneti, web sunucusunu, OSI modelini ve internet bağlantılarına sahip tüm cihazları hedef alan en yaygın ve tehlikeli saldırı türüdür. DDoS saldırısı, interneti sular altında bırakmaya ve gerçek kullanıcılar için kullanılamaz hale getirmeye çalışır. Bu çalışmada, DDoS saldırısını yenmek ve gelen trafikleri kontrol etmek için Defence of Flooding Attacks modeli önerilmiştir. Trafikleri tanımlamak, sınıflandırmak ve kontrol etmek için yazılım aracı kullanılmıştır. Önerilen modelin performansını test etmek ve değerlendirmek için CICDDoS 2019 veri seti kullanılmıştır. Elde edilen sonuçlara göre önerilen modelin ilgili çalışma ile karşılaştırılmasında %99.6 doğrulukla mükemmel sonuçlar elde ettiği görülmüştür.
Stetoskop olan hastanın muayene edilmesiarduıno cihazı kullanma
The stethoscope is an audio medical expedient for auscultation, or hearing to interior sounds of an physical or human body. Arduino is a physical programming platform consisting of an I / O (I / O) board and development environment that includes an implementation of the Processing / Wiring language. LSTM is an artificial recurrent neural network (RNN) architecture used in the field of deep learning. Unlike standard feed forward neural networks, LSTM has feedback links. It can process not only single data points (such as images) but also entire data strings (such as speech or video). In this study, new method presented for heart diseases detection by combining deep learning techniques based on optimization algorithm. In the first stage, LSTM applied to classify the heart signals that are obtained by Arduino. The BBO applied to enhance the performance of the LSTM by applied to reduce the error rate of the LSTM model. The presented method presented satisfactory results when compared with previous studies.
Yazılım geliştirme döngüleri için çoklu akıllı öneri motorlarının hibrit optimizasyonu
The primary purpose is to create a hybrid recommendation system approach to improve the performance of such systems. This recommendation system would typically be used to assign or suggest a small number of developers suitable for troubleshooting a bug report. Managing collections inside bug repositories is software developers' task to fix any particular bugs that have been identified. Bugs are often created, so the number of developers needed is high, so it's hard to decide how many to assign to specific tasks. This analysis aims better to understand the outcomes of the latest scientific methods. We also addressed developer prioritization and how it can be used to determine the assignment of a problem to a developer. We have studied two aspects: first, the selection of bug reports using hybrid machine learning methods, and modelling developer prioritization in the bug repository and supporting developer assignment tasks with our model. Second, we modelled the relevant objectives suggested by the developers' backgrounds based on proven knowledge and experience. The study focuses on two topers' experience with fixing bugs and developer rankings in the App Store. We've tried to take better assignments using developer prioritization in bug repositories, e.g., bug triage, severity identification and re-opened bug prediction. We examine the output of the model in a representative sample of bug repositories. The results show that the prioritization of developers' prioritization triage worker and allow the program to solve the bugs more effectively in support of the software support has been clarified. After the introduction, Chapter 2 relates the literature; chapter 3 is about the methodologies used in studies, the findings and explanation are described in this thesis, chapter 4, the following chapter is the description accompanied by references.
Sözleşme algılamanın uygulanmasımakine kullanılan IoT ağlarında sistem öğrenme
Since the last decade, Internet of Things (IoT - Internet of Things) solutions have been created and applied in different branches of society, such as solutions for transport and urban communication, for example. In this way, IoT is a new paradigm, which is composed of a global network of machines and devices capable of interacting with each other [4]. IoT changes the way some real-world problems can be modeled in the cyber environment, and this is due to the ability to list device units on the network to perform small tasks, and then group the results. In this work we have designed and implemented a system whose main purpose is to assist in the process of implementing this approach, a machine learning technique known as Active Learning was used. This technique was chosen in the context of this work, due to its potential to induce predictive models from databases with the lowest number of labels. Thus, a particle swarm algorithm PSO was used on the database, which consists of a supervised learning algorithm to validate the results obtained, a second predictive model was induced, which did not use active learning for sample selection.This model was used as a basis for comparative analysis of the model resulting from the proposal of this work.
Bulanık genetik algoritma ile kablosuz sensör ağında enerji tasarrufu
In this thesis, we describe the basic concepts related to the topic of the dissertation. Considering that in this dissertation, the optimization of energy consumption in wireless sensor networks has been done with the help of fuzzy and genetic algorithms of collective intelligence. In this thesis the genetic algorithm and fuzzy logic is investigated in the filed of the wireless sensor network. As simulation result shows the performance of genetic algorithm based on fuzzy logic is high life time. In this thesis, genetic algorithm and fuzzy logic are examined in the file of wireless sensor network. The simulation result showed that the performance of the fuzzy logic based genetic algorithm was better than the LEACH method
İstatistik makine öğrenimi ve değiştirilmiş adaboost sınıflayıcıyla yüz özelliklerinin tespiti
Classification modules can be used in multiple and diverse applications. The general objective of this work is the classification of facial features using statistical learning algorithms, this means being able to detect and classify the largest possible number of features that can be found on a face using only examples of images of these, without using a priori no information of the given characteristics. In the present work, detectors and classifiers of the characteristics that are considered most significant were developed, Excellent results were reported in mouth detection, with detection rates greater than 99% and errors comparable to the error from manual mouth marking. The lens sorter also obtained excellent results, with detection rates of 95% for databases with controlled environments and of the order of 90% for databases with uncontrolled environments. Beard and mustache classifiers after using the mouth detector obtained very good results, with a detection rate of over 95% in databases with uncontrolled environments.
NESNE TESPİTİ VE TANIMA DAHA HIZLI R-CNN KULLANARAK RGBD GÖRÜNTÜLERİ
The process of splitting the image into many segments can be called as image segmentation. The main aim of this process is to recognize the object in an image. Let's take an example a person wants to cross the road the first thing he'll do is he'll look at his right side first then left. If no vehicle is coming then the person will try to cross the road apart from that, this person will also check whether there is any electric poll or dog or is there any whole in the ground etc. The point is we're doing object detection first then we're crossing the road. Same thing we'll train our machine to do but it can't recognize directly the way humans do. We have to train the model first for that we're doing image segmentation. This step is typically used to recognize the objects or other relevant information.
İçerik tabanlı görüntü alma yüksek seviye semantik
This master's thesis focuses on Content Based Image Retrieval (CBIR) with high level semantics using image processing and deep learning techniques. The development of image edge smoothening system using Convolutional Neural Network (CNN) for CBIR optimization task is in hot pursuit, with special attention being given to the smoothening of all the edges of image. Given its high propensity to meta-size, going hand in hand with severe decreases in preservation rates, and the high inter-edge variability in image appearance, as well as a strong requirement on the training of the physician properly de-noising an image can be considered a daunting task. The purpose of this research thesis is to use a deep learning and image processing pipeline for content based image retrieval for the segmentation of edges in an image using with hybrid techniques in deep learning and imaging. The literature review of different papers was conducted with different imaging model architectures for CBIR. The CNN custom model was created for the task, and deep learning technique (CNN) was used with different levels of fine tuning of hybrid image processing techniques. Screening for high edge filter to identify edges at high accuracy has been under debate. In current discussion, the suggested smoothening procedure is bone density based including possible prescreening techniques for content based image retrieval and optimization on Corel-1000 dataset. Development of new prediction models and automated smoothening could contribute to general screening programs in the future. The custom deep learning model architectures were designed to represent different depths. The idea behind this is to analyze the effect of increasing representational capacity to the results and visualizations for all edges in an image. Additionally, deep learning CNN model was created to represent traditional automated image processing approach. Image processing has been used in some edging of images, producing good results for example in segmentation of image area structure and segmentation of image smoothening areas under consideration. The study also attempts to find solutions to practical deep learning challenges such as low training speed and lack of transparency with an accuracy of 98.47% absolutely. The imaging and deep learning pipelines are optimized in order to exploit the available parallelism using the MATLAB programming language with multiple tools under consideration.
Optimal yörünge planlaması tasarımı modifiye sürü kullanılan robot kol için algoritma
Much attention has been paid to improve, achieve the efficient performance of robot arm and to cope with the unlimited needs and demands of the worldwide industrial revolution that necessitate the availability of high productivity and accuracy. The finding of an optimal path possesses a significant role in guiding and attaining the accurate robot arm movements. In the current work, the optimal path and trajectory planning of two-link robot arm with 2-DOF in 2-D static known environment has been analyzed by proposing the Particle Swarm Optimization algorithm (PSO) and cubic polynomial equations. The primary function of modified PSO method is to generate the free-collision shortest path through searching within the free Cartesian space. The simulation results illustrate the efficiency of proposed methods performance in finding and determining the optimal path and trajectory even in various degree of environmental complexity. Moreover, the experimental work has been tested for practical physical validation in order to check the quality of the planned path practically and to analyze the difficulties of optimal practical path planning practically.
İlişki kuralı veri madenciliği tekniğiyle görüntü bölümlendirme
Smart grids are electric grids that are composed of multiple power sources and devices connected to each other to provide better reliability in power generation and power management, modern developments of the smart grid aim at either improving the control of power sources and loads connected to the smart grid by developing a specialized software/hardware, or by improving the communication within the parts of the smart grid and the central control. In this paper we aim at improving both sides of the smart grid system (communication and control), we propose a fuzzy logic-based controller for renewable energy and fossil fuel sources in a grid and an internet of things-based monitoring system which oversees the state of the smart grid, faults that occur in the grid, and how the fuzzy controller overcomes those faults, all in which provide an extra layer of support to the smart grid.