Theses supervised by Dr. Öğr. Üyesi Oğuz Karan
15 theses · Altınbaş University
MFF-LSTM: Çok Ölçekli Özellik Füzyon Tabanlı Uzun Kısa TasarlamaYalan Haber Tespit Sistemi için Farklı Özelliklere Sahip Dönem Belleği
The rise in the usage of media has led to a significant increase in the raise of false information, making it imperative to combat this issue and reduce our reliance on such unreliable sources. Fake news can mislead people, spread rumors, and even impact the positions of political leaders. Detecting fake news has become crucial in this digital era, with direct messaging platforms and social media playing a major role in its proliferation. Various innovative techniques have been suggested to determine fake news, making the endeavor both intriguing and challenging. Hence, this synopsis aims to develop the adaptive learning model with multiscale feature fusion for fake news detection. The proposed system constitutes "text collection, text pre-processing, feature extraction and detection". Initially, the text input is collected from the benchmark datasets, which is then followed by the text stage of pre-processing. Here, the pre-processed text is obtained that is fed into the feature extraction phases. The three feature extraction techniques such as "Bidirectional Encoder Representations from Transformers (BERT), Term Frequency-Inverse Document Frequency (TF-IDF) and GloVe Embedding" are employed to provide the feature set 1, 2 and 3. Finally, these resultant features are given to "Multiscale Feature Fusion based Long Short Term Memory (MFF-LSTM)", where the features are fused together in multiscale manner and detection is taken place by LSTM. Therefore, the system evaluation is done by considering the distinct measures and compared among traditional approaches. Hence, the recommended model attains the desired results to detect the fake news that helps to evade the exploration of any false information.
Borsa İstanbul'da faaliyet gösteren teknoloji şirketlerinde lojistik regresyon modeli ile finansal başarısızlık tahminlemesi
Finansal başarısızlık ile karşı karşıya kalan her şirketin sonu iflas ile sonuçlanmaz. Bununla beraber, global düzeyde etki gösteren ekonomik krizin oldukça fazla oranda şirketin tasfiye edilmesine ve iflas etmesine sebep olduğu aşikardır. Finansal başarısızlık kavramı konusunda görüş birliği ve net bir tanımlama yoktur. Bu da, finansal başarısızlığın tahmin edilmesine dair araştırmaların iflaslarla ilişkilendirilmesine sebebiyet vermektedir. Finansal başarısızlık, uzun senelerdir şirketler, yatırımcılar ve kredi verenler bakımından en kritik tehdit olarak görülmektedir. Küreselleşmenin son yıllardaki etkisiyle, teknolojik ilerlemelerin ve finansal koşulların da yansımasıyla ulusal ve uluslararası sektörlerin hızla farklılaşmakta ve ilerlemekte olduğu görülmektedir. Bu farklılık ve ilerleme, daha çok şirketler bakımından, finansal başarısızlıkları tahmin etmeyi ön plana çıkarmaktadır. Bu çalışmada finansal başarısızlık kavramı, finansal başarısızlık çeşitleri, finansal başarısızlığın önlenmesi konusunda alınabilecek tedbirler, finansal başarısızlığın tahmininde kullanılan yöntemler, finansal teknoloji, finansal teknoloji süreci, finansal teknoloji oyuncuları ve Borsa İstanbul kavramları kavramsal olarak anlatılmıştır. Çalışmada finansal veriler ile şirketlerin başarısızlıklarının lojistik regresyon yöntemi ile tahmin edilmesi amaçlanmıştır. Borsa İstanbul'da işlem gören 27 adet teknoloji şirketine ait veriler toplanarak rasyolar oluşturulmuştur. 2021 verileri ve son üç yıldaki finansal veriler ile başarısız şirketler belirlenmiştir. Şirketlere ait veriler öncelikle3 aylık dönemlere göre oluşturulmuş fakat bazı şirketlerin 2018 ve 2019 yıllarına ait sadece yıllık verilerine ulaşılabilmesi sonucunda yıllık veriler kullanılmasına karar verilmiştir. 2018, 2019 ve 2020 yıllarına ait belirlenen rasyolar bağımsız değişken olarak kullanılmıştır. Eksik verilerin bulunduğu rasyolar çıkarıldığında 60 farklı rasyo ile analiz yapılmıştır. Şirketlerin finansal başarısızlığının önceden tahmin edilmesinde 3 yıl önceden tahmin etmede Duran Varlıklar/Maddi Özkaynak, Net Borç/FAVÖK(Yıllık) ve Toplam Borç/Özsermaye rasyoları, 2 yıl önceden tahmin etmede FAVÖK/Büyüme (Yıllık) ve Kısa Vade Borç/Toplam Borç rasyoları ve 1 yıl önceden tahmin etmede Özsermaye Karlılığı/ROE ve Kısa Vade Borç/Büyüme rasyolarının başarılı olduğu sonucu ortaya çıkmıştır.
Web based GIS optimazation for coverage in wireless sensor network
In order to give flood forecasts and warnings, it is important to do research into the creation of a real-time GIS model that can be utilized for the visualization, modeling, and analysis of watershed management. This paradigm must be fully integrated with the Internet. To create an original prototype for a geographic information system (GIS), it is necessary to examine a diverse array of cutting-edge technologies. Due to these brand-new technologies, Internet-based computer modeling may now be carried out in a variety of distinct ways. Hydrology is researched in relation to the many technologies now available and their prospective uses. Using this framework, a number of geographical data structures may be included into a single conceptual framework that can accommodate a wide range of data models. To be useful for field monitoring data, it must have a rapid turnaround time and connect in real time with the geographic information system. As a result, it must be able to swiftly validate environmental simulation models and improve the accuracy of forecasts by analyzing hydrological flows and events in real time. As a last step, it should evaluate whether or not the explored concepts and skills are actual and interact with people
Identification of the speaker's voice using machine learning
The modern society of today has an ever-increasing demand for identity and safety measures, which has resulted in the development of a large number of different safety and identification measures. This demand is driving innovation in the field. Biometrics is a term that refers to many techniques that can be used to determine an individual's identity based on the unique characteristics that they have. The primary contribution that we have made with this paper is that we have designed and implemented a system that makes use of machine learning techniques in order to classify into labels features that have been collected from an audio file of a speaker. These features include the speaker's name, the speaker's gender, the speaker's age, and the speaker's location. The purpose of this research is to finally give a comprehensive understanding of the many approaches to Speech Recognition that are now accessible. This will be accomplished by performing a thorough examination of the relevant literature in order to achieve this goal.
A novel algorithm for selecting optimal routing paths to improve network performance
In this research, we explore various path methods utilizing Multiprotocol Label Switching (MPLS) to improve the term Traffic Engineering (TE) in networks. As internet services continue to grow in importance, efficient routing algorithms are crucial for TE. MPLS is a contemporary technique that improves router speed, identifies explicit routers, and facilitates better Traffic Engineering Management and real-time multimedia data communication. Among the protocols utilized to determine the shortest path for packet transmission in a network, MPLS is a significant one. However, ISPs must enhance the routing algorithm on MPLS routers to optimize their efficiency. Advanced routing algorithms have been developed to leverage MPLS networks, utilizing traffic-based routing to find label switching path (LSP) solutions and reduce blocking probability. To evaluate these algorithms, we utilized Matlab to simulate the MPLS network. The simulation results were analyzed and evaluated from various perspectives, and recommendations were made for specific routing algorithms based on the network specification protocol. The outcomes of this research indicate that integrating path methods for computer networks in terms of packet switching provides the best optimal route path
Node localization in WSN based on range measurement and reference locations
This research presents a decentralized method for addressing the difficulties associated with the n-dimensional static sensor network node localization. Consider a network in which n+1 nodes are anchored in place, but the remaining nodes are fully free-floating without any connections to other nodes. By applying barycentric coordinates obtained from range measurements, the non-convex node localization problem may be reduced in the noiseless situation into a conventional linear system of equations. This is accomplished by the use of barycentric coordinates. When independent zero mean Gaussian noise is introduced to range measurements, all barycentric coordinates become dependent random variables without a known standard distribution, and the distributions of these random variables may not be identical to one another. To solve the problem of noisy range-only localization, we provide a method that is based on online optimization techniques and a distributed online gradient descent algorithm.
Makine öğrenmesini kullanarak çevrimiçi reklamlara kullanıcı tıklamalarını tahmin etmek
This thesis explores the detection of user behavior within online advertising to better understand user preferences and refine ad strategies. As online ads are instrumental in reaching broad audiences, targeted approaches are vital for enhancing campaign effectiveness. The research employs data analysis, preprocessing, and machine learning (ML) techniques on a dataset detailing user behaviors like browsing history, ad clicks, and demographics. ML methods, including random forest, GB, and LR, are utilized alongside XAI tools like LIME and SHAP to highlight feature importance. The study aims to discern user behavior patterns to improve ad targeting. Models are further optimized using parameter tuning, and ensemble strategies, such as soft and hard voting, are used to aggregate individual predictions, boosting detection accuracy. Performance metrics, ranging from accuracy to specificity, are used to assess model efficacy. Results show that ensemble methods, particularly soft voting, outperform other techniques in accuracy.
İç mekan uygulamaları için hibrit metal-grafen mikroşerit anten tasarımı ve iyileştirilmesi
In recent years, wireless communications have gained significant prominence in our daily lives, offering a convenient and efficient means of transmitting information without the need for physical connections. This advancement has revolutionized the way tasks are performed and communication is carried out, facilitating seamless interactions over substantial distances using electromagnetic waves. The distinct advantage of wireless communication lies in its accessibility and adaptability, particularly in locations where conventional wired connections are impractical. As the reliance on wireless networks has grown, there has been a surge in connected devices and applications. However, this has led to network congestion, subsequently affecting data transmission speed. To counteract this challenge, network operators have explored solutions like utilizing higher frequencies to cater to the increasing demand for faster data transmission. The antenna plays a pivotal role within wireless communication systems, acting as a vital component. Among the various antenna types, microstrip antennas have gained prominence. Yet, they often exhibit limitations such as low gain and restricted bandwidth. Addressing these concerns becomes imperative. This study focuses on the enhancement of microstrip patch antennas (MPAs), specifically from rectangular and circular patches. The goal is to optimize their performance for operation at 62 GHz, catering to personal network applications within indoor environments. The simulation and optimization processes are carried out using CST Antennas modeling software. To enhance the bandwidth of the MPAs, a creative approach involving graphene flakes is employed. These flakes are strategically placed within engraved slots on the antenna patch. This introduces tunability to the antenna's frequency response, making it adaptable to applied voltages. Various slot shapes are explored to assess their impact on antenna performance. The simulation results highlight the effectiveness of this approach. The antennas exhibit robust performance and noteworthy tuning capabilities across multiple frequency bands, including 62 GHz, 59.545 GHz, 58.02 GHz, 53.21 GHz, and 57.965 GHz. This work underscores the potential of graphene-based modifications in enhancing antenna performance and expanding their applications in the realm of wireless communication. Keywords: Microstrip Antenna, Indoor Communications, Graphene, Reconfigurable Antenna, Antenna Optimization
Yangın ve duman algılama tabanlı yapay zeka teknikleri
An efficient monitoring system is required for accurate fire and smoke detection to stop the fire and guarantee the safety of the occupants' lives. This calls into question the existing sophisticated capabilities of fire alarm systems and highlights the need for a thorough smoke and fire detection system. It's critical to detect smoke and fire. Given the time and accuracy needed for fire detection, convolutional (deep) neural networks have been developed to recognize objects, even though flames usually inflict significant damage. YOLO was applied, and a suggested method called YOLOv8 was also created. Still, much research has been done on using real data with deep learning. The suggested method was to use an image-rich Smoke and Fire database. The findings show that the suggested approach performs better than others in accuracy, model size, and detection speed. Consequently, we demonstrated how to apply the YOLOv8l algorithm as quickly as possible to identify important fire and smoke properties; identification is faster and more accurate than earlier methods; the YOLOv8 algorithm achieves an average @mAP of 96.6%. Keywords: Artificial Intelligence, Convolutional Neural Network, Deep Learning, Image Processing, Fire and Smoke Detection, YOLOv8
UWB uygulamaları için ayarlanabilir bant aralığına sahip grafen tabanlı tek kutuplu mikroşerit anten
The utilization of Monopole Microstrip Antennas (M-MSAs) is widespread due to their affordability, simplicity in construction, and compact size, allowing seamless integration into various portable applications. Recently, the Ultrawide Band (UWB) technology, widely employed in wireless utilisations, is particularly reliant on this type of antenna. The UWB technology offers an extensive frequency range spanning in the range 3.1-10.6 GHz, catering to a multitude of small-power wireless implementations, including wireless voice transmission, personal localisation, radio frequency recognition, radar, and the HD video transmission. However, the broad nature of this frequency band increases the likelihood of interference. This contribution study seeks to formulate, simulate, and optimise an altered small-sized square M-MSA competent of fulfilling the conditions of UWB technology such as the large bandwidth. The introduced M-MSA design encompasses a square radiated patch, a dielectric material with a stout of 1 mm and 4.7 relative permittivity, a partially ground plane printed on the same face as the patch, and coplanar waveguide feed. To ensure the best compatibility with the UWB full band, certain of alterations to the M-MSA design are made like as a cut on the lower corners of the patch and changing the dimensions. Furthermore, the addition of a U-shaped aperture on the patch is suggested to be etched to create a bandgap within the UWB frequencies, effectively expected to reduce the interference. To obtain the tunability for the developed bandgap, graphene is employed to fill the aperture. In the case of the DC voltage being subjected to the graphene, the bandgap dissipates, while without biasing, the graphene exhibits high impedance, restricting current flow in the aperture. Consequently, the bandgap's impact becomes apparent within the frequency range of 3.87-4.85 GHz. After conducting the simulation and modification process, the obtained results demonstrate a significant improvement in both gain and efficiency. Furthermore, a notable increase in attenuation and degradation of gain is observed within the bandgap region, which was deliberately selected to mitigate interference from various sources such as military fixed communications, mobile communications, unmanned aerial vehicles, short-range radio links, satellite communications, and the low band of 5G. Keywords: 5G, M-MSA, Coplanar Waveguide, Graphene, UWB
Yoğun net kullanılarak bitki hastalıklarının tespiti
The precise and prompt identification of plant diseases (PD) is paramount in safeguarding crop productivity and fortifying global food security. As deep learning (DL) methodologies continue to evolve, the prominence of automated PD prediction mechanisms has witnessed a significant upswing. This study delves into the formulation of such an advanced system. Here, contemporary architectures, such as CNNs, DenseNet, ResNet, MobileNet, and VGG, are meticulously implemented, evaluated, and then juxtaposed against one another. Amidst these, DenseNet stands out, registering an exemplary ACC of 98%, thereby outstripping other state-of-the-art models. However, beyond mere ACC, our study extends to ensure the interpretability of the model's predictions. By integrating Explainable Artificial Intelligence (XAI) techniques, specifically the Grad-CAM approach, we strive to illuminate the decision-making pathways of the model, granting users a transparent view into its diagnostic process. This not only fortifies trust in the predictions but also provides valuable insights for potential real-world applications. In essence, this research underscores the pivotal role of automation in PD detection and demonstrates the compelling potential of blending DL with XAI for enhanced transparency and efficacy.
Yapay zeka tekniklerine dayanarak köprü yüzeyi çatlak tespiti
The manual inspection of concrete bridge surfaces for cracks takes a long time and wastes a lot of materials and labor. This study examines this problem by offering a quick and automated machine-learning method for visually inspecting concrete bridge surfaces with unmanned aerial vehicles (UAVs) in large open spaces. UAVs are used to capture high-resolution photographs of concrete bridge surfaces, providing thorough coverage of the inspection area. Utilizing the potent YOLO (You Only Look Once) technique, deep learning technology, a subset of machine learning, enables single-shot detectors. Use is made specifically of the YOLO v8 model, which was very accurate machine learning techniques were used to train. Since this model is based on Convolutional Neural Networks (CNNs) with the CSPDarknet-53 backbone structure, accurate and trustworthy results are guaranteed. The suggested model has outstanding performance in machine learning-based real-time crack detection. The Deep Learning model achieves remarkable precision and correctness in crack detection with an accuracy rating of 98.54%. The deep learning model can accurately detect cracks while limiting false positives, according to the F1 score, a measure of precision and recall balance, which reaches 97%. The robustness and accuracy of the machine learning model are demonstrated by the mean average precision (mAP), a statistic frequently employed in machine learning, which is calculated to be 97.90%. With a mAP50-95 of 83.11%, the model also achieves a high mAP across various Intersections over Union (IoU) thresholds. The recall rate reaches 94.66%, ensuring a high detection rate for true cracks, while the precision rate measures an astonishing 99.41%, minimizing false alarms. Additionally, the model runs at a phenomenal 95.24 frames per second (fps), which makes it possible to use machine-learning approaches for effective real-time crack detection. This speed suits it for practical applications since it enables prompt decision-making and intervention. Our study suggests a new, effective machine-learning method for instantly detecting cracks in concrete bridge surfaces. Utilizing image collecting from UAVs, With the help of the YOLO algorithm. Keywords: Crack detection, Machine learning, Deep Learning, Unmanned Aerial Vehicles (UAVs), Convolutional Neural Networks (CNNs), YOLO v8
Steganografi kullanarak bulutta güvenli veri depolamave görsel kriptografi
Due to the fact that digital images include sensitive data from a variety of sources, including human, medical, satellite, and biometric, among others, they necessitate the highest level of security. The images in question include a multitude of vulnerabilities. In this era of cloud computing and virtualization, securing data from hackers requires an effective method that employs a hybrid algorithm to preserve data from several sorts of hackers. Visual image encryption and LSB steganography are two essential security techniques for protecting sensitive images from unauthorized access and modification. Image encryption transforms the image data into an unintelligible form, while steganography embeds the image data in a cover image to conceal its existence. The study of the behavior of dynamic systems that are extremely sensitive to initial conditions is chaos theory. Chaotic systems are useful for image encryption and steganography due to their unpredictable and erratic behavior. This thesis proposes a new and improved system for color image encryption and steganography based on chaos theory. Using a novel chaotic-based encryption algorithm, the proposed approach uses the Lorenz system to generate a random key to encrypt the color image. The encrypted image is then embedded in a cover image using a novel chaotic-based steganography algorithm. The stego image is then transmitted or stored securely. The proposed system has several advantages over existing methods. First, it is more secure and resistant to attacks because it uses two chaotic systems, the Lorenz system, the Henon map, and the novel encryption and steganography algorithms. Second, it is more scalable vi and efficient due to using the Lorenz system, which is a relatively simple chaotic system. Third, it is more compatible with existing image and video formats. This thesis also provides a comprehensive performance evaluation of the proposed system. The experimental results show that the proposed approach is highly secure, scalable, and efficient. The proposed method is also compatible with existing image and video formats.
Access methods to patients medical records
Quick access to the medical record while maintaining confidentiality is one of the most important methods that must be followed in hospital emergency departments, taking into account that the information is correct and reliable. This research deals with ways to access the medical record while preserving the confidentiality of the information, knowing that there are many methods such as a user name and a password or via the health card. This paper proposes to exploit modern technology in the medical field and activate its use in this field, with a focus on identifying biometrics to access the health record with privacy policies. The study includes designing and creating a website, in terms of the technologies used, it contains PHP and C #, in addition to Arduino, and it also contains electronic components. In the first stage, the website and Arduino network are created with the fingerprint device, after which the device is tested to see if the system is operating at full capacity. The result of the general study system was a website to view the patient's file in emergencies with a fingerprint.
An adaptive transceiver design for duplex visible light communication
The Visible light communications (VLC) become an attractive subject for many researchers recently the reason why is that the VLC has many benefits that couldn't be observed in any of the RF communication systems; many of the researches have been done in the field of the VLC that focusing on how to reach to the high data rates for communication and handful of very little research gave interest on how to design a transceiver for VLC step by step to be utilised for short-range in an indoor environment application. In this study, a transceiver system design is proposed for a short distance and indoor communications that are utilise the VLC technique. The proposed system comprises two principal blocks (i.e. Block #1 and Block #2) each block is able to send the visible light data or receive the same data form. Block #1 comprises ATmega328P, Light Emitting Diode (LED), Liquid Crystal Display (LCD), and photo-sensor; this block is able to transmit the visible light data through the LED as well as receive the visible light data that is transmitted by the laser beam. While block #2 comprises ATmega328P, laser module, LCD, and LDR; this block is able to transmit the visible light data through the laser beam as well as receive the visible light data that is transmitted by the LED. Each block utilised the Ps2 keyboard to initialisation the data (i.e. numbers or letters) after that the Arduino microcontroller receive the character form the keyboard and transmit it as a bitstream via the LED or the laser beam. A good performance is achieved after the practical test applied and a communication range that is up to 30cm.