Department of Electrical & Electronics Engineering

Permanent URI for this collection

Browse

Recent Submissions

Now showing 1 - 5 of 81
  • Item
    PERFORMANCE EVALUATION OF PREPROCESSING TO PCA COMBINED MACHINE LEARNING TECHNIQUES ON PHARMACEUTICAL AND MINERAL SAMPLES BY LASER-INDUCED BREAKDOWN SPECTROSCOPY
    ( 2023-01-27) YAZICI, Göktuğ ; DORUK, Reşat Özgür
    For the purpose of identifying and analyzing materials, laser-induced breakdown spectroscopy (LIBS) is a quick optical nuclear discharge spectroscopy. It has the advantages of in-situ analysis, removal of rigorous sample processing, and micro-destructive properties for the substance being evaluated. LIBS uses brief bursts of laser beams to stimulate the material to a certain threshold, resulting in plasma formation. The plasma properties, which include wavelength value and intensity amplitude, are affected by the material and the surroundings of the experiment. The spectrum profiles of medication and mineral samples were obtained using LIBS in this study. The collection of pharmaceutical samples comprises two distinct concentrations of both paracetamol-based drugs, Aferin and Parafon. Aluminum (Al), Bizmut (Bi), Copper (Cu), Iron (Fe), Manganese (Mn), Nickel-Aluminum (NiAl), Tin (Sn), and Zinc (Zn) are among the mineral samples in the dataset. The samples' spectrum data were preprocessed by replacing missing values with shape-preserving piecewise cubic spline interpolation, filling outliers based on quartiles, smoothing spectra to remove noise, and normalizing both the wavelength and intensity axes. Statistical information was acquired, and both the preprocessed and raw datasets were subjected to principal component analysis (PCA). The machine learning models were built using two distinct train-test splits: 70% training - 30% test and 80% training - 20% test. Cross-validation was employed to keep the models from being overfit, hence the sample size is small. Both splits' machine learning outcomes from preprocessed and raw datasets were compared. This is the first time that all supervised machine learning classification algorithms, including Decision Trees, Discriminant, Nave Bayes, Support Vector Machines (SVM), k-NN (k-Nearest Neighbor), Ensemble Learning, and Neural Network algorithms, have been applied to LIBS datasets of both paracetamol-based pharmaceutical samples and 8 different mineral samples, as well as their preprocessed and raw datasets, to investigate the effect of preprocessing.
  • Item
    A STUDY ON MICROSTRIP ANTENNA DESIGN FOR 77 GHZ RADAR SYSTEMS
    ( 2023-01-20) YILMAZ, Selen ; KARA, Ali ; DALVEREN, Yaser
    This thesis presents a comprehensive investigation into the design and operational behavior of series-fed microstrip patch antenna array for the 77 GHz automotive radar. Initially, the theoretical background information on the theory of microstrip antenna, patch antenna array, frequency scanning array and Chebyshev array are provided. A full-wave finite element method-based simulation tool is used to design and slightly tune the dimensions of the antennas as a parametric study. At the first stage, a series fed linear Chebyshev patch array with resonance at 76.5 GHz is designed representing one transmit channel of the antenna. Shorting pins are loaded to transition structure of ground-signal-ground (GSG) padding to enhance the total gain. Comparative analysis between vialess and via loaded cases is conducted in terms of bandwidth and gain. At the last stage, 76.5 GHz linear patch antenna array is converted into a 79 GHz linear patch antenna array by optimizing the GSG padding dimensions, scaling the spacings between each two adjacent array elements and the length of array elements. Two designs are proposed to assess the effect of scaling method at this stage. Comparative analysis in terms of the beam steering angle, the impedance bandwidth, the overall gain and the sidelobe level suppression is conducted between these two designs. Keywords: Dolph-Chebyshev Distribution, Frequency Scanning Array, Linear Array, mmWave, Patch Array Antenna.
  • Item
    MEDICAL DATASET CLASSIFICATION BASED ON DIFFERENT DEEP LEARNING TECHNIQUES AND META-HEURISTIC ALGORITHMS
    ( 2023-01-30) KADHIM, Yezi Ali ; MISHRA, Alok ; DORUK, Reşat Özgür
    Medicine is one of the fields where computer science advancement is making significant progress. The usage of Computers in Medical improves precision and accelerates data processing and diagnosis. There are currently a variety of computer assisted diagnostic systems, with deep-learning algorithms playing an important role. Systems that are more precise and faster are required. Computer-aided diagnosis (CAD) has proved to be an effective and accurate method for diagnostic prediction over the years. This study focuses on the development of an automated CAD system with the intent to perform diagnosis as accurately as possible. Deep learning methods have been able to produce impressive results on medical image datasets. In this thesis, several combinations of different deep learning techniques with the meta-heuristic algorithm, each of the deep learning methods either the convolutional neural network or the auto-encoder were applied to extract the effective features from two different medical datasets COVID-19, and the brain tumor with the aid of the meta-heuristic method to select the optimal features in order to cover a several medical datasets detections. The first combination of several pre-trained convolutional neural networks (CNN) AlexNet, GoogleNet, ResNet 50, and DenseNet 201 was used with three types of Meta-Heuristic Algorithms Ant Colony Optimization algorithm (ACO), Particle Swarm Optimization algorithm (PSO), and Genetic Algorithm (GA). The second combination was Auto-encoder with three types Meta-Heuristic Algorithms ACO, PSO, and GA which was an innovative method, that seeks to reduce the size of the dataset while maintaining the original performance of the data. The employing of deep learning methods in conjunction with meta-heuristic algorithms and supervised machine-learning algorithms to perform an accurate diagnosis. Pre-trained convolutional neural networks (CNNs) or auto-encoder are used for feature extraction, whereas feature selection is performed using ACO or PSO, or GA. Ant colony optimization helps to search for the best optimal features while reducing the amount of data. Lastly, diagnosis prediction (classification) is achieved using learnable classifiers. The novel framework for the extraction and selection of features is based on deep learning, auto-encoder, and ACO. The performance of the proposed combination is evaluated with classifiers like decision tree (DT), support vector machine (SVM), k-nearest neighbors (KNN), ensemble, Naive Bayes, and discriminant using two medical image datasets: chest X-ray (CXR) and magnetic resonance imaging (MRI) for the prediction of the existence of COVID-19 and brain tumors. Accuracy is used as the main measure to compare the performance of the proposed approach with existing state-of-the-art methods. The proposed system achieves an average accuracy of 99.61% and 99.18%, outperforming all other methods in diagnosing the presence of COVID-19 and brain tumors, respectively. Based on the achieved results, it can be claimed that physicians or radiologists can confidently utilize the proposed approach for diagnosing COVID-19 patients and patients with specific brain tumors. Also, in this thesis, a combination of different deep learning techniques with the meta-heuristic algorithm, each of the deep learning methods either the convolutional neural network or the auto-encoder were applied to extract the effective features from two different medical datasets, with the aid of the meta heuristic method to select the optimal features obtained by the particle swarm optimization algorithm (PSO) this combination is considered as an innovative method, seeks to reduce the size of the dataset while maintaining the original performance of the data. The covid-19 dataset found that the highest accuracy by the combination of CNN-PSO-SVM was 99.76%, and for the common brain tumor dataset, the accuracy of 99.51% as the highest was obtained by the combination method autoencoder-PSO KNN. We notice that the combination model of the deep learning method with the PSO feature selection algorithm takes a consuming time much longer than the same method with the ACO algorithm at the same time the accuracy of PSO is near to ACO accuracy.
  • Item
    MEDICAL DATASET CLASSIFICATION BASED ON DIFFERENT DEEP LEARNING TECHNIQUES AND META-HEURISTIC ALGORITHMS
    ( 2023-01-30) KADHIM, Yezi Ali ; MISHRA, Alok ; DORUK, Reşat Özgür
    Medicine is one of the fields where computer science advancement is making significant progress. The usage of Computers in Medical improves precision and accelerates data processing and diagnosis. There are currently a variety of computer assisted diagnostic systems, with deep-learning algorithms playing an important role. Systems that are more precise and faster are required. Computer-aided diagnosis (CAD) has proved to be an effective and accurate method for diagnostic prediction over the years. This study focuses on the development of an automated CAD system with the intent to perform diagnosis as accurately as possible. Deep learning methods have been able to produce impressive results on medical image datasets. In this thesis, several combinations of different deep learning techniques with the meta-heuristic algorithm, each of the deep learning methods either the convolutional neural network or the auto-encoder were applied to extract the effective features from two different medical datasets COVID-19, and the brain tumor with the aid of the meta-heuristic method to select the optimal features in order to cover a several medical datasets detections. The first combination of several pre-trained convolutional neural networks (CNN) AlexNet, GoogleNet, ResNet 50, and DenseNet 201 was used with three types of Meta-Heuristic Algorithms Ant Colony Optimization algorithm (ACO), Particle Swarm Optimization algorithm (PSO), and Genetic Algorithm (GA). The second combination was Auto-encoder with three types Meta-Heuristic Algorithms ACO, PSO, and GA which was an innovative method, that seeks to reduce the size of the dataset while maintaining the original performance of the data. The employing of deep learning methods in conjunction with meta-heuristic algorithms and supervised machine-learning algorithms to perform an accurate diagnosis. Pre-trained convolutional neural networks (CNNs) or auto-encoder are used for feature extraction, whereas feature selection is performed using ACO or PSO, or GA. Ant colony optimization helps to search for the best optimal features while reducing the amount of data. Lastly, diagnosis prediction (classification) is achieved using learnable classifiers. The novel framework for the extraction and selection of features is based on deep learning, auto-encoder, and ACO. The performance of the proposed combination is evaluated with classifiers like decision tree (DT), support vector machine (SVM), k-nearest neighbors (KNN), ensemble, Naive Bayes, and discriminant using two medical image datasets: chest X-ray (CXR) and magnetic resonance imaging (MRI) for the prediction of the existence of COVID-19 and brain tumors. Accuracy is used as the main measure to compare the performance of the proposed approach with existing state-of-the-art methods. The proposed system achieves an average accuracy of 99.61% and 99.18%, outperforming all other methods in diagnosing the presence of COVID-19 and brain tumors, respectively. Based on the achieved results, it can be claimed that physicians or radiologists can confidently utilize the proposed approach for diagnosing COVID-19 patients and patients with specific brain tumors. Also, in this thesis, a combination of different deep learning techniques with the meta-heuristic algorithm, each of the deep learning methods either the convolutional neural network or the auto-encoder were applied to extract the effective features from two different medical datasets, with the aid of the meta heuristic method to select the optimal features obtained by the particle swarm optimization algorithm (PSO) this combination is considered as an innovative method, seeks to reduce the size of the dataset while maintaining the original performance of the data. The covid-19 dataset found that the highest accuracy by the combination of CNN-PSO-SVM was 99.76%, and for the common brain tumor dataset, the accuracy of 99.51% as the highest was obtained by the combination method autoencoder-PSO KNN. We notice that the combination model of the deep learning method with the PSO feature selection algorithm takes a consuming time much longer than the same method with the ACO algorithm at the same time the accuracy of PSO is near to ACO accuracy.
  • Item
    COMPARATIVE ANALYSIS OF MPPT TECHNIQUES FOR SOLAR AND WIND SYSTEMS UNDER DIFFERENT OPERATING CONDITIONS
    ( 2022-12-26) AHMAD, Muhammad Saeed ; SÜNTER, Sedat
    Renewable energy technologies have gained a lot of traction in the last few decades as a means of reducing reliance on fossil fuels and mitigating the impact of climate change. Renewable sources such as sunlight, wind, and water are clean and sustainable. These technologies have gained significant attention in recent years. While renewable energy technologies have many advantages, one of the main challenges is their relatively low efficiency compared to fossil fuels. As a result, renewable energy systems typically require more land and resources to produce the same amount of energy as fossil fuel-based systems. Additionally, the efficiency of renewable energy systems can vary depending on the weather and other environmental conditions. For example, solar panels are less effective on cloudy days and wind turbines are less effective in calm weather. This can make it difficult to predict and control the amount of energy that renewable systems will produce, which can create challenges for integrating them into the grid. The problem with efficiency can be dealt with the use of maximum power point tracking (MPPT) techniques. These techniques are used to optimize the performance of renewable energy systems by ensuring that they operate at the maximum power point, or the point at which they can generate the most power. There are several types of maximum power point tracking (MPPT) techniques, but they can be broadly classified into three categories: simple, artificial intelligence (AI), and hybrid. Simple MPPT techniques such as PO and IC are the most basic and widely used type of MPPT. These techniques use relatively simple algorithms to continuously adjust the operating conditions of the system to maintain the maximum power point. AI-based MPPT techniques like PSO and ANN use advanced algorithms and machine learning techniques to optimize the performance of renewable energy systems. These techniques can adapt to changing environmental conditions and can continuously adjust the operating conditions of the system in real-time. Hybrid MPPT techniques like ANFIS and PSO&PO are a combination of simple and AI based techniques. These techniques use simple algorithms to quickly track the maximum power point, and then use AI-based techniques to fine-tune the operating conditions of the system in real-time. A comparative analysis of simple, AI, ML, and hybrid MPPT techniques for hybrid energy (Solar and Wind) systems is discussed in this thesis. The MPPT algorithms were ranked based on different metrics such as efficiency, settling time, oscillations at MPPT and algorithm complexity. For PV system, AI based techniques performed best as compared to Hybrid and conventional techniques. For Wind system, hybrid techniques yield the best results as they combine the benefits of conventional and AI techniques.