Theses supervised by Dr. Öğr. Üyesi Abdullahı Abdu Ibrahım
36 theses · Altınbaş University
Advanced data analytics for network security: Detecting and mitigating threats through real-time data processing
This research aims to investigate the enhancement of real-time threat identification and mitigation through the application of advanced data analytics in network security. The findings of this literature review and qualitative interviews with field specialists indicate that real-time data processing significantly improves the precision of threat detection and the velocity of response. We highlight advanced machine learning techniques such as decision trees and neural networks due to their ability to identify patterns overlooked by traditional methods. The research underscores the essential requirement for a workforce skilled in data analytics and cybersecurity, as well as the significance of a systematic implementation approach. Recommendations for organizations encompass employee training, utilization of real-time data, and the establishment of systematic deployment strategies. This study has certain limitations, such as the exclusive use of qualitative data. Nonetheless, it demonstrates that data analytics can enhance network security management. It recommends that subsequent research examine ethical issues, such as data protection and privacy within advanced analytics, and do empirical assessments of the proposed methodologies. Businesses must modify their security protocols to leverage data analytics capacity to mitigate the effects of emerging cyber threats.
The role of machine learning in enhancing realism in unreal engine games
The way machine learning (ML) is applied in game production process, especially through development IDE such as the Unreal Engine 5, paves innovative methods for developing significant immersive and authentic digital experiences. This thesis inspects the relational intersection of machine learning and game development process with examination of spatial how machine learning methods lends to gameplay mechanics, player interactions, and simulates physics-based movements. Additionally this thesis also contains a case study about movement mechanics of wind turbines in a simulated environment, using Unreal Engine's blueprint scripting way of programming but also hardcode in C++ to make blueprints run into Unreal Engine's blueprint scripting. Used coded simulation using real time Machine Learning algorithm on C++ to calculate the turbine's body and blade movement and simulated it with real world factors such as wind source, distance and speed. This calculation impacts the amount of power that is produced by the wind turbine. This tells that applying these methods and technologies with all of ML algorithms, Unreal Engine 5's in engine physics simulations and time may enhance visual environments, fidelity and deep game environment mechanics. This work demonstrates that ML based simulations can be done in real time in a way that reflects player behaviors and environmental influences in the game thereby improving the gameplay experience. The game utilizes AI-driven simulations that respond to player behavior, as well as dynamic interactions with environmental changes, contributing to greater immersion. Finally, this research explores finds and includes the challenges for importing real time Machine Learning algorithms into games, uncovering the potential resource usage and performance optimizations, leads and provides insights into to potential future of advanced ML and AI powered game designs, mechanics and coding. This thesis and project provides an improvement in knowledge on using the ML and AI algorithms into the gaming part of life and innovative thinking of adapting these technologies for realism techniques to extend the limit of your in-game interactions and not Static experiences in virtual fronts.
Ethical implications of ai driven personalization in digital marketing
The digital marketing industry is undergoing a transformation owing to the use of machine learning with Artificial intelligence (AI). This technology permits the industry to provide a highly personalized experience for consumers. As soon as it derives to AI, there's considerable discourse about engagement and devotion. The purpose is simple: AI bids a way to distinguish "stuff" at rate never before conceivable. Hitherto, despite the advent of these glowing outcomes, numerous ethical queries need answering. These ethical features have been inspected through thorough study of the literature and case studies to attain a modernized sorting of noticeable issues. These are: (1) the irresistible quantity of data being gathered and the confidentiality risks required; (2) the prejudices that algorithms can symbolize and the discrimination they can preserve; (3) the likely for consumer influence; (4) the economic disruption that AI might cause; (5) the transparency issues and the accountability shortfall that it generates. Although AI-powered personalization is able to produce additional pertinent and tempting consumer familiarities, it also presents serious ethical predicaments that must be addressed to guarantee responsible and evenhanded applications. The study concludes through recommendations for vendors and legislators to resolve the remunerations of AI personalization with the inevitability for ethical morals and consumer fortification, ultimately aiming to respect human principles and improve inclusive well-being.
Investigation, simulation and improvement of energy reduction algorithms in wireless sensor networks
Today, due to the advantages of wireless sensor networks, which are simple and inexpensive implementation, low power consumption and high scalability, they have been used in many applications. Designing stable wireless sensor networks is a very challenging issue. Sensors with limited energy are expected to operate automatically for a long time. However, replacing broken batteries may be costly or even difficult in harsh environments. On the other hand, unlike other networks, wireless sensor networks are designed for specific small-scale applications such as medical surveillance systems and large-scale applications such as environmental monitoring. In this regard, a great deal of research has been done to propose a wide range of solutions to the problem of energy saving. In this paper, a routing algorithm is designed to produce the best path between the sensor nodes and the local collector node with the aim of achieving a proper traffic distribution and thus creating a balance in the energy consumption of the intermediate nodes. Creating such a balance will help increase the life of the network and improve the energy consumption pattern of wireless sensor networks with limited energy resources. On the other hand, by using the possibility of changing the range of nodes, it is tried to increase the possibility of load distribution in low-density points of the network. The simulation results show a 20% improvement in network life using the proposed algorithm compared to some of the proposed energy-sensitive routing algorithms in recent years
Fog and cloud load balancing using regression based recurrent deep learning algorithm
Fog computing has been studied by a variety of academics, and they have identified concerns which need to be solved. It is the primary goal of this project to build and execute an energy conscious load balancer that will reduce energy usage and help in the distribution of loads. Furthermore, with chips becoming more compact and containing increasingly dense circuitry, the release of energy in the form of heat increases. This means even more energy consumption, as computational components do not operate well at very high temperatures and therefore are cooling systems. In this work we investigate the relative strengths and weaknesses of the different fog and cloud models and structures
Optimal routing in internet of things networks using artificial intelligence
IoT will touch society more than past digital revolutions. It impacts global business, environment, and safety. IoT won't replace humans. Billion-connected Things require expertise. IoT enhances living, resource usage, and industry efficiency. If security precautions are not followed, the quantity of networked devices makes them vulnerable to intrusion, which can entail financial loss and death. Guard devices. Cybersecurity-certified devices are released. IoT creates commercial prospects. Benefits require infrastructure-to-endpoint security. Secure people and processes. Human less security wastes millions. Users voted this. Trusting devices and weak passwords. Most corporate users support weak-pass worded third-party devices on the internal network. "Every company will be invaded, we simply don't know when," said a consultant and strategic adviser. Home and corporate users should understand device vulnerabilities and risks before using them. AI, 5G, and edge computing improve IoT. We proposed sensor-coupled module chains and simulated annealing module placement for edge computing. Our simulation reduced energy and latency. Cloud and IoT are new ideas. Communication, distributed cognition, and on-demand massively parallel processing enable it. Cloud computing helps IoT. Complex IoT demands more computing. Understanding computer activity scheduling optimization approaches is vital. In a multi-cloud system with poor parallelism, enormous arrival rates, and variable runtimes, we investigated simulated annealing. Discrete event simulators evaluated system performance and cost. Simulations show this scheduling method increases performance and cost.
Speaker voice recognition using feature selection and SVM classification
Gender recognition based solely on the speaker's voice is a fairly simple task for any human being, however, it's not as simple as it is for humans compared to any computing systems, the task requires multiple tedious processes of feature recognition and selection, and multiple computational processes to acquire such experience in gender recognition. neural networks (NN) has always been the best choice when it comes to image and audio classification and other pattern analysis tasks, the accuracy and precision of the output results widened the prospect of utilizing the different ANN variations in voice recognition, in this paper we present a system design for using K-nearest neighbor network to further enhance the results of the gender detection results by the voice recognition process.
A new deep learning-based framework for cyberscurity problems
In this study, we proposes new study based CNN-GA-random forest to detect the SQL injection attacks in IoTs. In the first stage, the CNN applied to extract high level features from input SQL inquiries. Then, the output of the CNN wired to the random forest. The random forest is robust classifier used in several classification and regression problems and presented remarkable results when compared with other classifiers. Then, the genetic algorithm applied to train the CNN to select best weight and basis of the model. The genetic algorithm is robust optimization algorithm and used in several fields to enhance the performance of the models such as design, classification, regression and estimation. The proposed system showed results with an accuracy of 99.93% compared to some studies.
Detecting denial of service attack of network traffic by build accurate intrusion detection system based on machine learning algorithm
Today, the creation of more effective intrusion detection systems has become crucial due to the rise in computer malware. Ensure the availability of the system is an important component of information security and the most important requirement of any network. Recently the Machine Learning algorithm (ML) has been used to improve intrusion detection over the network. It is currently necessary to release an updated version of these systems. The presented work aimed to build a reliable and accurate IDS based on ML to classify and prevent distributed denial of service attacks to protect any system working on the network from temporary or complete system failure. The proposed ML models, including (decision tree, random forest, logistic regression, support vector machine, and multi-layer neural network) were trained and evaluated using the cic-ids-2018 dataset. Furthermore, principal component analysis (PCA) was used to reduce the dimensionality of the dataset. According to the classification results, the proposed multi-layer neural network model has optimal performance at an accuracy of 99.9992%.
A novel logistic transfer using optimization algorithms in natural disasters and humanitarian services
Today, there are thousands of organizations involved in the activities of transporting goods, services or people from one point to another. Due to the dynamic conditions of our era, the goals and limitations of organizations are highly variable. In this context, one of the most common problems faced by organizations is the impossibility of carrying out activities such as the distribution of vehicles to service points, transportation, and logistics. In this study, efficient Logistic transfer algorithm presented to find minimum path by using PSO based Ant Conny optimization method. The main contribution in this study is combining PSO with Ant Conny optimization method to find the best path with minimum execution time. Several scenarios are executed to validate the presented method. Finally, the obtained results compared with several studies presented to solve the same problem.
Design and analysis of microstrip antenna with frequency selective surface superstrate
Recently, wireless communications have significantly advanced in our daily lives, particularly in terms of our ability to perform tasks and communicate without a physical media. The ultimate crucial aspect of via distance communication is the capacity to transmit data over long distances using electromagnetic waves. These systems are simple to install and operate in locations where sending physical media is challenging. As the number of appliances and implementations linked to wireless networks has increased, congestion has occurred, slowing data transmission. Switching to high frequencies can be considered as one of the practical ways introduced by the network developers to satisfy users' demand for quick data transmission. In wireless communication architecture, the receiver or transmitter antenna is among the ultimate essential elements. Whereas, microstrip antennas are one type of antenna that is utilized, but it is necessary to increase these parameters due to their restricted bandwidth and depressed gain. In this thesis, we will introduce a design for regular MPA for the utilisation of 5G based 28 GHz. To obtain the optimum outcomes from the MPA we introduce two scenarios for the purpose of performance improvement. The first one is by employing the Frequency Selective Surface (FSS) that will be fitted over the MPA with a determined air gap, whereas the FSS will only pass the required frequency and reject the others. This will improve the gain but will degrade the bandwidth. While the second one is the utilisation of the array structure that will boost the total performance of the antenna. On the basis of the obtained outcomes, the use of array arrangement enhances the gain, bandwidth, and return loss. On the other hand, the power loss is increased due to the feeding network.
Design and optimization of a state-of the-art solar pv system relying on maximum power point tracking of solar charge controller using artificial neural networks
The total effectiveness of the PV system is significantly impacted by the use of an effective MPPT (Maximum Power Point Tracking) algorithm. These algorithms are utilized as tracking controllers to get the most power possible out of PV modules based on the array temperature, solar radiation, shading circumstances, and PV cell ageing. (P&O) Perturb and observe and (InC) incremental conductance are the most used approaches. These time-tested methods are inexpensive, easy to use, and of modest efficiency. MPPT techniques using artificial intelligence for better steady-state and transient performance include fuzzy logic and artificial neural network (ANN) controllers. A proposed artificial neural network-based model of a solar power tracking system is what this thesis aims to create. Through this study, a single PV system using a Buck DC-DC Converter is tracked for maximum power under various irradiation situations employing the P&O and ANN approaches. To be able to analyse the PV array's model, determine MPP, and show the results, this study employs MATLAB-Simulink. Buck converter is the type which is used to study the converter DC/DC performance. The findings show that the ANN model can track changes in MPP far more effectively than the approach of Perturb and Observe. The system using ANN appears to deliver higher amount of power than the model using P&O since the maximum power monitored by ANN is 1990 Watts while the maximum power monitored by P&O is still 1930 Watts. Voltage and current ripples are greatly reduced by the ANN model compared to the P&O model. Therefore, ANN responds more quickly than P&O. Additionally, the use of Buck converter, offers us a better voltage, power, and thus, efficiency. The authors propose further research on novel strategies that can be efficient and useful, where MPPT methods additionally take into consideration external repercussions without concern for the cost and complexity of sensing.
Tekstil sektöründe perakende satış analizi ve tahminlemesi
This thesis addresses sales forecasting in the retail domain using machine learning methods. Nowadays, the retail sector emerges as a rapidly changing and exponentially growing field. Therefore, obtaining accurate and reliable predictions in sales forecasting holds critical importance for businesses to sustain their competitive advantage. Machine learning is recognized as an effective tool with data analysis and pattern recognition capabilities to address complex problems like sales forecasting. This study aims to tackle the sales forecasting problem in the retail sector and investigate how machine learning methods can be utilized to solve this issue. Firstly, an exploration of the existing sales forecasting methods in the literature and machine learning algorithms is conducted. Subsequently, the effectiveness and performance of various machine learning algorithms (such as LGBM, LSTM, XGBoost) in sales forecasting are compared. This thesis is supported by experimental studies conducted on real-world datasets. The datasets encompass sales data from the textile retail sector and comprise data obtained over a specific time frame. Analyses conducted on these datasets reveal how machine learning algorithms can offer an advantage in sales forecasting compared to traditional methods. The results demonstrate that machine learning algorithms constitute an effective and accurate tool for sales forecasting in the retail sector. This thesis provides valuable insights into the LightGBM method specifically chosen and can aid businesses in enhancing strategic decisions, such as demand management, inventory optimization, and stock planning.
IoT based smart greenhouse monitoring and control system with early flood detection
The aim of this thesis is to develop an internet of things based smart greenhouse with flood detection monitoring system that can be deployed using the existing WiFi networks, avoiding hard-installation processes and allowing users to add or remove devices from the system according to their needs using the simulated environment for automation as well as the greenhouse monitoring. The research took advantage of Internet-of-Things (IoT) based Message Queuing Telemetry Transport (MQTT) protocol to guarantee detection and interoperability in our system, such as the QoS levels that we use to specify the flood data, where according to the defined levels, the flood data could be processed with early forecast delivery assurance achieving an accuracy of 97.36%. Thus, the objective of this work is to present a solution for a smart greenhouse system that only uses the WiFi network for communication with MQTT protocol, avoiding intrusive installations (it does not need to add physical wiring, just effective python programming based advance simulation), and allows to connect and control different devices, from any vendor, remotely from a mobile phone, a tablet or any other device with an internet connection where the SEN-12 dataset has been used for flood data providing different geographical regions for potential flood. A complex and modular system has been developed, which uses standard technologies and protocols and consists in a central server (smart greenhouse server), which offers a set of services to control and manage the system; an auxiliary database to minimize information that devices need to store; a MQTT broker that allows the devices to communicate using this protocol, the MQTT, and finally the devices whose requirements are minimum, they only must able to connect to an WiFi network. The system does not limit the quantity of connected devices since its architecture was designed and implemented to allow horizontal scalability and high availability to handle with a growing amount of devices using the flood detection with low tariff of unit consumption
Design application of network management by using API (application programming interface)
Our daily lives depend heavily on devices, services and applications connected to the Internet. This reliance makes it more important than ever that the primary networks on which they depend are reliable, efficient and secure. At the same time, the increasing sophistication and diversity of devices, existing services and applications have made network management tasks more complex than ever before. Modern network management assumes that agents, general distributors of Internet networks can develop networks management by billing system in terms of many aspects and benefits on a regular basis, and use this information to take preventive or corrective actions in real time (control of network). Traditional approaches to network management In creating, configuring internet packages and distributing them its sometimes manual, slow need time, effort and negativity that are not appropriate for the desired performance measurements or applications. The contribution of this research is to design an application we have been call MaxBox, where it creates and configures internet networks and packets with the lowest possible time and less complicated, over three protocols Hotspot, broadband and User manager Through the application programming interface API, which in turn is a point of contact between a specific development environment and developers without the need to build everything from scratch. The purpose of the API is to hide the programming and method of write codes, such as command line interface CLI, then by the API is control the mikrotik
Video object tracking with artificial neural network and artificial bee colony optimization method
Video object tracking is one of the many problems in the field of Computer Vision; it is a basic component for many and more complex vision systems that are useful for several real world applications in the areas of medical research, surveillance, robotics, tele collaboration, etc. A video object tracking algorithm tries to follow an object of interest trough the frames of a given video sequence. This thesis studies the effects of performing video object tracking aided by the Honeybee Search Algorithm, a Swarm Intelligence (SI) algorithm that is inspired in the foraging behavior of honeybees; and Graphics Processing Units (GPUs), which are an example of a Parallel Computing technology designed specifically for graphic rendering operations. The main contribution is to develop and investigate the video based object tracking with the help of artificial neural network and artificial bee colony optimization method. This research intent to train the proposed system with artificial neural network on any open source dataset of video objects. However, for the methodology of proposed system, artificial bee colony would be utilized. Artificial bee colony optimization method is one of the most advance and efficient method of getting better results by combination of different agents represented as bee to perform any task on set of rules. The artificial neural network is already given in the machine learning toolbox of MATLAB. Dataset for video based objects are obtained from open source repository like Kaggle UCL and OpenSets. For the implementation of this thesis, MATLAB 2019a would be forwarded in use as it is the most efficient tool for processing large scale data with more accuracy. The development of said implementation and the demonstration that the Parallel Honeybee Search Algorithm can successfully be used to improve the time cost of a video object tracking algorithm in given situations, with little effect on the accuracy of the results. The results prove that it is possible to parallelize the Honeybee Search Algorithm and use it for video object tracking. In comparison with a parallel version of the same video object tracking algorithm, the addition of the Honeybee Search Algorithm helps to provide a more stable time to deliver results, making them less dependent on the size of the specific video, and without causing notable negative effects in the accuracy of the results.
Brain tumor detection and classification using image processing techniques
Brain Tumor is the abnormal and uncontrolled growth of tissues or cells in the brain, the brain tumor is dangerous and life-threatening disease, Detection of the diseases through image processing is done by using an integrated approach working methods of processing MRI images of brain tumor entering it and distinguishes this approach if the brain is normal or abnormal, computer systems have been used in this area to analyze medical information, analyzing and extracting the most important features of the brain tumor and focusing on image analysis and processing techniques to distinguish between different diseases based on the symptoms of each disease. This work adopts two proposed approaches for detecting brain tumor using image processing and deep learning techniques with makes a comparison between these two approaches. This work was planned to some an important and common group of brain tumor, including Glioma, Meningioma, Pituitary Adenoma, and Nerve Sheath. These kinds of brain tumors are the most popular in the world. The dataset contains 3000 images related to malignant (normal) and benign (abnormal) each one has 1500 image. In the first proposed approach, where several steps are used in the form of stages, which are include, the image acquisition stage, image pre-processing, image segmentation, image post-processing, extraction the features, and the classification stage. Class support vector machine (SVM) algorithm was used to perform the classification process in the second proposed approach, the convolution neural network (CNN) was used through which the brain tumor are classified according to a special structure of this algorithm consisting of several layers. In these two proposed approaches, the tumor were classified to deferent classes was detected. The obtained results from the comparison between the two proposed approaches in terms of performance and accuracy showed the preference of the second approach which adopted the deep learning and using the CNN algorithm, over the first approach, because the overall accuracy rate that obtained from the second proposed approach was (98,29%). While the overall accuracy rate that obtained from the first proposed approach was (68.9%). So, the second proposed approach is more accurate and powerful in the process of detecting and classifying brain tumor.
Hybrid detection techniques for skin cancer images
According to W.H.O, skin cancer is one of the most common types of human malignancy in medical sector. The development of skin cancer detection system using CNN based on skin lesion diagnostic tasks is in hot pursuit, with special attention being given to the classification of benign and malignant. Given its high propensity to meta-size, going hand in hand with severe decreases in survival rates, and the high inter-patient variability in lesion appearance, as well as a strong requirement on the training of the physician/dermatologist properly diagnosing a melanoma can be considered a daunting task. The purpose of this research thesis is to use a deep learning for the detection and segmentation of skin diseases using with hybrid techniques in deep learning and imaging. The literature review of different papers was conducted with different data mining model architectures. Three custom models were created for the task, and deep learning techniques were used with different levels of fine tuning of hybrid deep learning models. Screening for skin cancer diseases to identify people at high risk has been under debate. In current discussion, the suggested screening procedure is bone density based including possible prescreening questionnaires. Development of new prediction models and automated diagnostics 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 patients of skin cancer. Additionally, deep learning models were created to represent traditional data mining approach. Data mining has been used in some skin diseases, producing good results for example in segmentation of skin defected area structure and diagnosis of skin diseases. The study also attempts to find solutions to practical deep learning challenges such as low training speed and lack of transparency. There are two forms of skin tissue.
Evrimsel sinir ağı (CNN) kullanarak insan faaliyetlerinin tanıma için etkili bir model
Information can be obtained from individuals through the systems of classifying and recognizing human activities at any time. These systems are used in different areas such as the detection of diseases, improvement of physical therapy stages, development of smart home projects. In this study, the data obtained from accelerometer and gyroscope sensors on smart phones are used. Most of the studies in the literature are unable to analyse higher-level features and their relationships based on machine learning and deep learning techniques. Convolutional Neural Network (CNN) model is a very suitable deep learning approach due to its ability to obtain high level and sensitive features. The deep learning-based approach that includes this background has been used in the classification of various human activities in the experiments in our study. In the experiments, the classification performance accuracy rate was measured by giving different input parameters, layer and network units to the relevant network models. As a result, it has been shown that six different classes are classified with high accuracy, achieving a classification performance of approximately 97.98%
Speaker voice recognition using a hybrid PSO/fuzzy logic system
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.
Isolated word detection using LSTM based feature extraction methods
In this study, pitch applied as feature extraction techniques and combined with RNN which these two methods are new approaches which PSO applied as RNN trainer. The Pitch based RNN presented remarkable results, which extracted features by pitch, wired to RNN and classified to seven words and presented 97.12% accuracy. The aim of applying PSO is to optimize the RNN and which find best weights and basis of the model. The presented framework presented best results than previous researches in the field of speech recognition.
Terör ve suç faaliyetleriyle mücadele için geliştirilmiş genetik algoritma kullanarak verimli bir İHA yolu planlama yaklaşımı
Drones are a very popular type of aircraft that has seemingly taken over the world. Other widely used names for drones are Unmanned Aerial Vehicles (UAVs) and Remotely Piloted Aerial Systems (RPAS). Drones are being extensively consumed in today's world, not just by government authorities and law enforcement but also by commercial and private entities. However, the effectiveness of using drones as a long-term strategy for counter-terrorism, counter-insurgency, controlling criminal activities remain unclear. This research aims to examine the perceived effectiveness of using drones in combating terrorism and criminal activities. Moreover, presenting new path planning techniques based on tree seed algorithm and genetic algorithm (TSAGA). The presented approach applied to find the best path for the drone when used to visit more than one location or goal. The proposed TSAGA assist the drones to find the minimum path which lead to Energy saving and solving the energy problem in the drones which is one of the main problems in drone technology. Then, the obtained results by the TSAGA compared with classical GA and show that the proposed approach is best and better in both energy saving and execution time.
Fourier transform based epileptic seizure features classification using scalp electrical measurements using KNN and SVM
The great development that took place in the technology of the interaction between humans and computers led to remarkable and incredible success in many scientific fields. Until this day, researchers' studies are continuing in this field to reach the highest possible accuracy to obtain the results that approximate the accuracy of human work. Electroencephalogram (EEG) is one of the devices that took a wide space and led many studies to amazing results in the field of recording, analyzing, detecting, and classifying brain signals. Where this technology was able to monitor the disorders, which happened in the brain and provide the ability to study the state of health of the brain. In addition, machine learning had Various techniques that were successfully involved in the classification of EEG signals. Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) were our specialties here among the most suitable techniques for classifying EEG data. This thesis aims to build a system to assist clinicians on three levels, first assisting in patients' automatic monitoring which leads to a digital memory easy to memorize. Secondly, reducing the work time by Supporting the clinicians' decision which assists acceleration in getting the goals with the least number of errors. Finally, assisting in identifying the appropriate medication and medical care that patients may need. Thus, for achieving this purpose, the dataset used here was from Temple University Hospital Seizure Corpus (TUH) [1]. TUH is considered one of the largest open-source databases and is the most extensive [2-4]. Moreover, to ensure the best results preprocessing techniques were implemented to Dataset. There are many techniques for EEG signal processing that feed the classification models with the best data. Fast Fourier Transform (FFT) was studied as one of the types of feature extraction methods for processing EEG signals. Eventually, the results associated with classifying seizure types showed SVM got the best classification accuracy compared with KNN where the accuracy was 99.5 % and 99 %, respectively
Geliştirilmiş AI tekniğini kullanan yazılım hatası tahmini için yeni çerçeve
In this thesis, new framework planned for software defect estimation using LSTM using TSA. The proposed framework combined AI technique with TSA to optimize the defect predication accuracy. Then, in the first stage the software defect dataset become input to the PCA. Furthermore, the TSA applied to optimize the bias and weight of the LSTM. The proposed framework validated by using number of datasets proposed by NASA datasets to improve the framework results. The experimental results presented by using by calculating several parameters to evaluate the proposed method. Additionally, the obtained results compared with various studies proposed in this field.