Artificial Intelligence Studies https://aistudies.org/index.php/ais <p>Artificial Intelligence Studies (AIS) is an International Refereed Journal published in any field releated to Artificial Intelligence. The objective of AIS is to be able to hear scientific studies both at&nbsp; international academic and industrial organizations.</p> Parantez Teknoloji San. Tic. Ltd. Şti. en-US Artificial Intelligence Studies 2651-5350 <p>Artificial Intelligence Studies (AIS) publishes open access articles under a Creative Commons Attribution 4.0 International License (CC BY). This license permits user to freely share (copy, distribute and transmit) and adapt the contribution including for commercial purposes, as long as the author is properly attributed.</p> <p><img style="width: 88px;" src="https://licensebuttons.net/l/by/4.0/88x31.png" alt="88x31.png"></p> <p><strong style="font-size: .9em;">&nbsp;</strong><strong>For all licenses mentioned above, authors can retain copyright and all publication rights without restriction.&nbsp;</strong></p> <p><strong>&nbsp;</strong></p> <p>&nbsp;</p> Classification of Breast Ultrasound Images Based on Regional and Morphological Features https://aistudies.org/index.php/ais/article/view/67 <p>Abstract – Breast cancer is a highly prevalent and the most lethal cancer type in women, emphasizing the critical importance of early diagnosis and treatment. This study is based on extracting features from breast ultrasound images (BUSI) from a publicly available dataset. The research examined types of breast cancer based on regional and morphological features extracted from mask images of breast ultrasound.</p> <p>Regional and morphological features were extracted from BUS images, and the least absolute shrinkage and selection operator (LASSO) method was used for feature selection. The results demonstrated that the selected features could effectively distinguish between malignant and benign breast lesions with high accuracy. In this study, machine learning methods such as support vector machines (SVM), artificial neural networks (ANN), and naive bayes (NB) were employed to classify benign and malignant lesions. The classification methods were evaluated using various performance criteria. According to the results, in the study conducted with balanced data, the ANN method achieved the highest classification performance, with an area under the curve (AUC) value of 0.9973 and an accuracy value of 0.9887.</p> Zeynep Ak Şerife Gengeç Benli ##submission.copyrightStatement## http://creativecommons.org/licenses/by-nc/4.0 2024-12-29 2024-12-29 7 2 28 35 10.30855/AIS.2024.07.02.01 Classification of Lung and Colon Cancer Histopathological Images using a Novel Artificial Intelligence Method https://aistudies.org/index.php/ais/article/view/72 <p>Cancer is a disease in which cells acquire autonomous growth, genetic instability, and significant metastatic strength, and is considered one of the most common causes of death worldwide. The most important types of cancer-causing these deaths are lung and colon cancers. Although they are rarely seen at the same time, the rate of metastasis of cancerous cells between these two organs is quite high if not diagnosed early. Histopathological diagnosis and appropriate treatment are the only ways to distinguish cancer types and reduce cancer death rates. The use of artificial intelligence in histopathological diagnosis can also provide experts with significant assistance with less effort, time, and cost.</p> <p>In this study a dataset, containing 25000 histopathological images belonging to 5 classes to classify colon and lung cancer types, was used. In order to obtain successful classification results from this dataset, the versions of the DenseNet algorithm, one of the deep learning algorithms, (DenseNet121, DenseNet169, and DenseNet201) were used firstly. Then, 3 novel models (DenseNet121_Improved, DenseNet169_Improved, and DenseNet201_Improved) were proposed by adding a cut-point layer, an auxiliary layer, and making frozen status improvements to the versions of the DenseNet algorithm. Versions of the DenseNet algorithm and proposed models were trained with stratified k-fold cross-validation technique first on colon cancer containing 2-class histopathological images, then lung cancer containing 3-class histopathological images, and lastly on 5-class histopathological images containing both colon and lung cancer. Finally, classification success rates were obtained. According to the experimental results performed on 3 different datasets, 97.60%, and 98.48% classification success rates in the lung cancer dataset and in both colon and lung cancer datasets were obtained respectively. The best classification success rate was achieved with DenseNet201_Improved, which was recommended with 99.80% in the colon cancer dataset.</p> İsmail Akgül Volkan Kaya Tuğba Bal Taştan ##submission.copyrightStatement## http://creativecommons.org/licenses/by-nc/4.0 2024-12-29 2024-12-29 7 2 36 52 10.30855/AIS.2024.07.02.02 Use of Machine Learning and Deep Learning Algorithms the Diagnosis of Thyroid Nodules https://aistudies.org/index.php/ais/article/view/75 <p>This research explores the effectiveness of machine learning and deep learning algorithms in evaluating thyroid nodules, with a particular focus on their ability to assist in making diagnostic decisions. Experimental studies evaluate the success of different models in classifying thyroid nodules.</p> <p>The results include a detailed analysis based on Positive Predictive Value (PPV) and False Discovery Rate (FDR) values on benign, malignant and normal classes. Model-1 showed high performance in the malignant and normal classes, although there is potential for improvement in the benign class. Model-2 improved the performance in the benign and normal classes, but there is still room for improvement in the benign class.</p> <p>In particular, Model-3 and Model-4 showed a successful classification performance, achieving high PPV values in all three classes. These models stand out with high PPV and low FDR values, especially in the benign class. These findings highlight the significant promise of deep learning algorithms in improving the assessment of thyroid nodules, playing a key role in improving diagnostic accuracy and clinical decision making.</p> <p>However, considering the areas for improvement, it seems that more studies and improvements are needed, especially in the benign class. Future research should focus on the use of larger datasets and improved methodologies to improve model performance and use it more reliably in clinical applications.</p> Süleyman Menderes İsmail Şahin ##submission.copyrightStatement## http://creativecommons.org/licenses/by-nc/4.0 2024-12-29 2024-12-29 7 2 53 65 10.30855/AIS.2024.07.02.03 Digitization of Printed Multiple Choice Questions Using Object Detection Methods: A Yolov7-Based Approach https://aistudies.org/index.php/ais/article/view/76 <p>In education, multiple-choice questions are widely used in exams. Hard copies of multiple-choice questions are usually used to prepare students for exams. However, when hard-copy questions are intended to be used in electronic environments, difficulties are experienced because editing cannot be done, and statistical and mathematical operations cannot be performed with hard-copy questions. This study uses object detection methods to transfer hard-copy multiple-choice questions to digital media. Within the scope of the study, a new approach is presented to determine the locations of multiple-choice questions and convert them to text using the YOLOv7 algorithm. There are 16 different class labels, such as question stem, question number, and question options, in the proposed YOLO model. During the data preprocessing phase, 1,140 images were manually labeled using the Roboflow image data labeling system, resulting in the creation of a unique dataset. As a result of model training, it was seen that the model classified approximately 94% of the questions correctly. The data detected with the developed model was converted to XML format using the algorithms of the YOLOv7 library. Educational institutions can use this approach to extract data and understand visual information.</p> Yiğit Çetinkaya Adem Tekerek ##submission.copyrightStatement## http://creativecommons.org/licenses/by-nc/4.0 2024-12-29 2024-12-29 7 2 66 82 10.30855/AIS.2024.07.02.04 A Deep Ensemble Reinforcement Learning Based Approach For Stock Trading https://aistudies.org/index.php/ais/article/view/78 <p><span style="vertical-align: inherit;"><span style="vertical-align: inherit;">Algoritmik ticaret, kantitatif ticaret olarak da bilinir, finans ve teknoloji endüstrilerinde önemli bir rol oynar. Borsa yatırımcılarının alım satım işlemlerinde kullandıkları bazı analitik yöntemlerin yerini alan algoritmik ticaret, makine öğrenimi ve derin öğrenmedeki gelişmelerden yararlanarak karmaşık verilerden anlam çıkarmada derin öğrenme yöntemlerinin yeteneğinden yararlanır. Bu çalışmada, topluluk öğrenmesi adı verilen bir çerçeve, derin öğrenme yöntemi olan takviyeli öğrenme ile birleştirildi. Eğitilmiş takviyeli öğrenme aracısı, Standard &amp; Poors 500 (GSPC) endeksinin 2011 verileri üzerinde test edildi ve 2.258,27 $ kar marjı elde edildi. Önerilen yapıda, Uzun Kısa Süreli Bellek (LSTM) aracısı zaman serisi verilerini işlerken, Evrişimli Sinir Ağı (CNN) aracısı bu verilerden oluşturulan bir görüntüyü girdi olarak kullandı. Bu girdilerden elde edilen tahminler birleştirildi ve nihai sonuç bir Derin Q Ağı (DQN) modelinden türetildi, böylece topluluk öğrenme yapısı oluşturuldu. Sonuçlar, literatürdeki benzer çalışmalarda, topluluk öğrenme yöntemi kullanılarak yapılan tahminlerin, ajanlar ve yöntemler tarafından yapılan bireysel tahminlere kıyasla daha yüksek kazançlar sağladığını göstermektedir.</span></span></p> Seyfullah Arslan Durmuş Özdemir ##submission.copyrightStatement## http://creativecommons.org/licenses/by-nc/4.0 2024-12-29 2024-12-29 7 2 83 96 10.30855/AIS.2024.07.02.05 Comparative Assessment of Machine Learning Models in EEG-Based Sentence Classification https://aistudies.org/index.php/ais/article/view/79 <p>In this study conducted a detailed analysis of widely used machine learning algorithms for sentence classification on TSEEG (Turkish EEG) dataset, examining both classification accuracy and computational efficiency. Given the complex nature of EEG signals, the study also investigates the necessity and impact of preprocessing techniques—such as normalization, filtering, and feature extraction—on classification performance. Among the models evaluated, the Support Vector Machine (SVM) achieved the highest test accuracy (99.17%) and F1 score (99.12%), demonstrating strong reliability and processing speed. To further validate the reliability and robustness of the SVM model, cross-validation was used, confirming its stability across different data subsets. This study not only emphasizes the critical role of preprocessing in enhancing EEG data analysis but also provides practical benchmarks on accuracy and processing time, offering valuable guidance for model selection in similar research. By identifying effective algorithmic choices, this work supports future research in making informed decisions regarding preprocessing and model selection across diverse EEG datasets and application domains.</p> Emrullah Şahin Durmuş Özdemir ##submission.copyrightStatement## http://creativecommons.org/licenses/by-nc/4.0 2024-12-29 2024-12-29 7 2 97 114 10.30855/AIS.2024.07.02.06 LoRa-Enabled IoT-Based Smart Irrigation Systems: Water Resource Management and Efficiency https://aistudies.org/index.php/ais/article/view/80 <p>Increasing population, climate change, and global warming are causing the world’s available water resources to decrease rapidly. Therefore, it is vital to use existing water resources efficiently. Developing technology offers promising solutions to overcome this issue. In this study, the design of the Internet of Things-based smart irrigation system, one of today's popular technologies, is presented. The presented system consists of IoT nodes divided into four different categories which are central node, irrigation nodes, sensor nodes, and control nodes. The central node, which controls all irrigation processes, collects data from irrigation and control nodes via LoRa communication modules. Sensor nodes transmit humidity information from their respective cultivation areas via Wi-Fi using the MQTT communication protocol. Irrigation nodes pre-process the data coming from the sensor nodes and transmit it to the central node, they also control the water valves going to the area to be irrigated and monitor the internal pressures of the water pipes. The control node, another node in the system, constantly monitors the water level of the water tank and manages the activation of water pumps. The user interface software which has a direct connection with the central node allows users to create, monitor, and update irrigation schedules. LoRa technology ensures reliable long-distance data transmission and enables the system to operate cost-effectively and energy-efficiently. Aditionally, The proposed system saves water and energy while optimizing irrigation needs according to plant types.</p> Fırat Aydemir ##submission.copyrightStatement## http://creativecommons.org/licenses/by-nc/4.0 2024-12-29 2024-12-29 7 2 115 119 10.30855/AIS.2024.07.02.07 Performance Analysis of Deep Learning Architectures in Brain Tumor Classification https://aistudies.org/index.php/ais/article/view/81 <p>Brain tumors are uncontrolled tissue growths that develop as a result of replication errors during the cell renewal process and are classified as malignant or benign. These tumors can cause brain damage or severe health problems due to increased intracranial pressure. Generally, brain tumors are categorized into two main types: primary and secondary. Primary tumors, often benign, originate directly in the brain, whereas secondary tumors result from the metastasis of cancer cells from other organs to the brain. Early diagnosis of brain tumors is crucial for treatment planning and evaluating outcomes. This study aims to facilitate the early detection and classification of brain tumors using Magnetic Resonance Imaging (MRI). It evaluates the performance of deep learning architectures such as EfficientNet, DenseNet, and AlexNet on three different datasets containing glioma, meningioma, pituitary tumors, and tumor-free brain images. The research compares current deep learning techniques and various performance metrics to identify architectures with the highest performance. The results demonstrate that the developed methods are effective in brain tumor classification. This research aims to assist healthcare professionals in making accurate diagnoses, enabling swift and effective treatment of patients, and significantly contributing to the early diagnosis and treatment planning of brain tumors.</p> Çetin Erçelik Ali Arı ##submission.copyrightStatement## http://creativecommons.org/licenses/by-nc/4.0 2024-12-29 2024-12-29 7 2 120 130 10.30855/AIS.2024.07.02.08