Rojiar Pir mohammadiani

Agriculture is essential for sustaining human life. As the global population is expected to reach 10 billion by the mid-21st century, ensuring food security presents significant challenges. Traditional agricultural practices, which have historically met the dietary needs of the population, may no longer be sufficient to support such a large number of individuals. Modern agriculture enhances productivity by integrating IoT and machine learning technologies. In recent years, Iraq has experienced significant climate changes, reducing the availability of groundwater crucial for irrigation. Despite a long-standing water agreement with Turkey, Iraq continues to face water scarcity issues. This research demonstrates that implementing intelligent irrigation systems can conserve water and enhance agricultural productivity in the region. Although research shows that 61% of farming studies focus on crop management, less than 10% address irrigation strategies. Effective irrigation management, however, significantly influences crop yields. In our approach, we manage the irrigation of various crops, including strawberries, vegetables, and tomatoes, using IoT-enabled devices and sensors such as temperature, humidity, light intensity, and irrigation sensors. Devices such as Arduino Uno and Ethernet Shield collect data and transmit it to an edge server for processing. During our research, we engineered an advanced irrigation system tailored to various crops. This system employs machine learning techniques, specifically multi-class classification algorithms, to create a sophisticated irrigation schedule that optimizes water usage across different types of crops. By integrating these cutting-edge technologies, our study aims to enhance agricultural efficiency and resource management, By using machine learning algorithms such as Random Forest, Support Vector Machines, Logistic Regression, and KNN, we can predict irrigation needs with an accuracy exceeding 95%. This data-driven strategy allows us to create precise irrigation schedules, improving both irrigation management and crop yields. The edge server sends data to a local web server and the ThingSpeak cloud.

Detecting communities in a graph means identifying substructures or subgroups of nodes that have close connections within themselves and less connections with other nodes in the graph. In many real networks, nodes have the possibility to be members of different communities, and in this way, the problem of recognizing overlapping communities in networks and finding appropriate centers and members of each community is still one of the challenges in this field, finding appropriate centers It is very important to try to select the centers automatically and to ensure that the members of each community do not have outlier data problems. The overlapping issue has also been investigated.In order to identify the communities and structures in the graph, this research includes three main steps: converting the graph into axial distance vectors, determining the important centers, and finally clustering the members based on their distance from the centers. In the first step, the deep walking method was used to convert the graph nodes into vectors. The second step involves using the Laplacian criterion to determine the centers of the clusters. In the third stage, which is clustering, members are assigned to the closest centers according to the distance from each center, so that they are related to the closest center.To evaluate the proposed method, four different data sets were used, namely football, poolbook dolphin, karate, and emails. Our proposed model for football, pool book, dolphin, karate, e-mail was able to achieve the highest accuracy by obtaining the accuracy of 0.95, 0.95, 1, 1 and 0.98. Also, in the NMI criterion with values of 0.98, 0. 74, 1, 1 and 0.86 as the best performance.

Breast cancer is one of the most common cancers among women, which endangers the patient's life if not diagnosed and predicted on time. A variety of medical imaging methods as well as biopsies help doctors to diagnose breast cancer. Since biopsy is invasive, therefor using medical imaging methods is safe for patients. A physician can diagnose the presence of a tumor by examining breast images. Computer science used in medical image processing and disease diagnosis. Deep learning is a subset of artificial intelligence that has achieved promising results in processing all kinds of images, especially medical images. In this research a proposed convolution neural network, a proposed canvolutioanl auto-encored, a pre-trained ResNet-50, a trained from scratch ResNet-50, a pre-trained Inception v3 and a trained from scratch Inception v3 were developed to classify ultrasound breast images. There are two scenarios for images classification. In the first scenario, images were classified into two classes of benign cancer and malignant cancer. In the second images were classified into three classes of benign cancer, malignant cancer and healthy control. The dataset contains 780 images in three classes: healthy or normal (133 images), malignant cancer (210 images) and benign cancer (487 images). The number of images increased to 6,413 images (including 2,185 benign cancer images, 2,100 malignant cancer images, and 2,128 healthy images) using data augmentation methods such as 5 degrees rotating, flipping horizontally, and flipping vertically. In the next step, the size of the images was 150*125 pixeles. The highest accuracy of 97% was obtained by pre-trained Inception V3 to classify images into two classes. The highest precision of 100% was obtained by trained from scratch ResNet-50 in benign calss. Pre-trained Inception v3 achieved the precision of 98% in malignant cancer calss. For sensetivity in benign calss and malignant class, the highest value of 98% and 99% were obtained by from scratch trained Inception V3 and in common canvolutioanl auto-encored and trained from scratch ResNet-50, respectively. The highest F1 score of 99% and 98% were obtained by the pre-trained Inception v3 in both classes of benign and malignant. For three classes classification, the pre-trained Inception v3 achived the highest accuracy of 96%. Also, the highest precision, senstivity and F1 score of 100%, 96% and 98% were obtained by pre-trained Inception v3 in benign class. In healty control group, the highest F1 score of 98%, 99% and 98% was obtained by convolution neural network, convolution neural network and pre-trained Inception v3.

Clustering attributed graphs, which involves learning node representations from node attributes and topological graph structures, is a fundamental and challenging task in the analysis of network-structured data. However, existing methods often overlook the distinctions between topological and non-topological information, resulting in redundant representations. To address this issue, this thesis introduces the Diverse Joint Nonnegative Matrix Tri-Factorization (Div-JNMTF), an embedding-based model that aims to uncover communities in attributed graphs. The novel Div-JNMTF model utilizes a joint nonnegative matrix tri-factorization approach to extract diverse node representations from both topological and non-topological data. To reduce redundancy among the representations and encourage the distinct contributions of each type of information, a diversity regularization technique is employed using the Hilbert-Schmidt Independence Criterion (HSIC). Additionally, two graph regularization terms are introduced to preserve the local structures in both the topological and attribute representation spaces. To solve the problem, an iterative optimization strategy is devised for the proposed method. Extensive experiments are conducted on eight attributed graph datasets, demonstrating the effectiveness of the Div-JNMTF model in accurately identifying attributed communities. The results indicate that it outperforms state-of-the-art methods in this domain.

In today's era, cryptocurrencies use blockchain technology in a decentralized manner for security and verification of transaction registration. It is one of the most discussed topics in this field. New currencies are also extracted in the process of recording transactions. With the difference that their construction is different from other digital currencies related to the government and basically this money belongs to the people. Bitcoin is the first and flagship cryptocurrency that has eliminated the biggest problem of double spending. Various methods of earning money in the cryptocurrency market without the need to buy and sell, including mining digital currencies and setting up a full node, building and buying and selling NFT, blockchain games, lending and cultivating profits, receiving currency airdrops Digitalization is staking or staking of digital currencies and investment and long-term purchase and maintenance of digital currencies. There are three important types of analysis in this market, which include technical analysis, fundamental analysis, and intra-chain analysis. In this thesis, we examine and analyze the technical trends and price patterns. By developing the relative strength index indicator algorithm and finding commonly used patterns such as triangles, flags, corners, doubles and doubles in the price chart, we have achieved a more reliable definition of entry and exit points. In this regard, we provide two tools (indicators) to help the trader in identifying more reliable entry and exit points. In a better look at the relative strength index, instead of the entry point of 30, we use 55 and instead of the exit point of 70, we use the points 55-70-80-90. We will also change the basis of closing 14 candles to two numbers 9 and 17. Also, the divergences and convergences are determined by the indicator itself. The experiments conducted by robots show an increase of up to 30 times in the detection of the more accurate trend of the relative power index.

امروزه وبسایتهای شبکه های اجتماعی به یک منبع غنی از داده های ناهمگون مبدل شده است؛ ازاینرو تجزیه و تحلیل این دادهها میتواند منجر به کشف اطلاعات و روابط ناشناخته در این یک چالش مهم درزمینهی تجزیهوتحلیل دادههای » مشابه « شبکه ها شود. کشف جامعه متشکل از گره های شبکهه ای اجتماعی است، و بهطور گستردهای درزمینهی ساختار گرافی در این شبکهها موردمطالعه قرارگرفته است. شبکههای اجتماعی اینترنتی علاوه بر ساختار گرافی، حاوی اطلاعات مفیدی از کاربران درون شبکه میباشند؛ که استفاده از این اطلاعات میتواند منجر به بهبود کیفت کشف جوامع گردد. در این پایاننامه، برای تشخیص جوامع، از اطلاعات ارتباطی و اطلاعات محتوایی استفادهشده است. در این روش ابتدا با کاوش الگوی تکرارشونده، الگوهای پرتکرار را براساس عملیات کاربران پیدا میکند و جوامع کوچکی را تشکیل میدهد که هم ازنظر ساختاری و هم ازنظر عملیات مشابه باشند، سپس با انتشار برچسب، هر جامعه را با استفاده از ارتباطات اجتماعی و اطلاعات محتوایی گسترش میدهیم.

Choosing the optimal set of influential people has become an attractive problem in complex networks. This problem is broken into two sub-problems: (1) finding the influential nodes and ranking them based on the individual influence of each node (2) finding a group of nodes to achieve the maximum spread in the network. In this thesis, both sub-problems have been examined and a method for measuring the spread power of influential nodes in the network and selecting the optimal group from them has been presented. In the proposed method, first the input network is divided into different communities. Then, the edges of each community are weighted and in each of the communities, the spreading power of the nodes is measured and ranked. Finally, a group of influential nodes were selected to start the publishing process. Data sets of real networks have been used to evaluate the methods. The proposed method was compared with other previously known methods in two parts. In the first part, the accuracy of the method in measuring the spread power of network nodes is compared based on the resolution and similarity parameters, and in the second part, the proposed method is compared with other methods in terms of the spread amount of influence of the selected set. The obtained results show the significant superiority of the proposed method in all three evaluation criteria over other methods.

پیش بینی پیوند در شبکه های پیچیده یکی از موضوعات ضروری در حوزه داده کاوی و کشف دانش در چند سال گذشته بوده است. در واقع؛ این مسأله به دنبال پاسخ این سؤال است که اگر تصویر لحظه ی کنونی شبکه در اختیار باشد، احتمالاً چه روابط جدیدی میان موجودیت های شبکه شکل خواهد گرفت. در این بین روش های مبتنی بر شباهت به دلیل سادگی و عملکرد مناسب از محبوب ترین روش های پیش بینی پیوند محسوب می شوند. هدف اصلی این پایان نامه، بهبود دقت روش های مبتنی برشباهت پیش بینی پیوند با استفاده از اطلاعات جوامع می باشد. اطلاعات مورد استفاده در این پژوهش برگرفته از ساختار گراف و مبتنی بر شبیه ترین نود ها در جوامع مشترک در یک نمایش سلسله مراتبی است که منجر به معرفی یک معیار جدید شده است. این معیار از تعداد شبیه ترین نود ها در جوامع مشترک بین دو رأس به دست می آید. در این معیار هر چقدر تعداد نود های با اهمیت از نظربیشترین شباهت در جوامع مشترک بیشتر باشد، آنگاه این دو رأس با احتمال بیشتری امکان تشکیل یال خواهند داشت. برای ارزیابی روش های ارائه شده از دو مجموعه داده شبکه واقعی بیولوژیکی از جمله شبکه زیست شناسی و عصبی و سه شبکه فیس بوک، نویسندگی مشترک دانشمندان و پیوند های میان وبلاگ و درنهایت یک مجموعه داده شبکه مصنوعی استفاده می شود. برای تست این روش از چهار الگوریتم پیش بینی پیوند پایه مبتنی بر همسایگی از قبیل همسایه های مشترک(CN)، ضریب جاکارد(JC)، تخصیص منابع(RA)، آدامیک-آدار(AA) و برای ارزیابی آن از معیار های AUC و Precision استفاده می شود. نتایج نشان می دهند که استفاده از تعداد شبیه ترین نود ها در جوامع مشترک بین دو رأس، با برخورداری از یک پیچیدگی زمانی مناسب، در رابطه با بیشتر مجموعه داده ها منجر به بهبود دقت در پیش بینی پیوند خواهد شد.

با گسترش سریع تکنولوژی اطلاعات، داده ها عموماً با تعداد ویژگی های زیادی در بسیاری از حوزه ها ظاهر می شوند. این داده ها نه تنها پیچیدگی های محاسباتی و نیازهای حافظه ای الگوریتم-های یادگیری را افزایش می دهند، بلکه عملکرد آن ها را نیز بدتر می کنند؛ به دلیل وجود ویژگی-های غیرمرتبط، افزونه و اختلالی. کاهش ابعاد ویژگی فرایند انتخاب یک زیر مجموعه از ویژگی هایی است که حاوی اطلاعات مفید برای ایجاد مدل هستند، و در الگوریتم های یادگیری ماشین، روشی برای افزایش سرعت الگوریتم و غلبه بر بیش برازش است. در این پایان نامه تمرکز بر روی انتخاب ویژگی از نوع بدون نظارت است که به دلیل نبود برچسب داده ها مسئله چالش برانگیزی است، و روش جدیدی برای انتخاب ویژگی از نوع بدون نظارت ارائه می شود. در روش پیشنهادی، داده ها ی ورودی فاقد برچسب فرض شده اند که این روش در روش پیشنهادی اول از رمزگذار-رمزگشا استفاده می کند؛ به نحوی که از رمزگذار برای تبدیل داده های اصلی به بازنمایی با ابعاد پایین و هم زمان از رمزگشا برای بازسازی داده های اصلی به کمک همان بازنمایی ابعاد پایین استفاده می کند که با این روش نتایج به نسبت برخی روش های مطرح بهبود پیدا کرده اند و در روش پیشنهادی دوم دوم برای بهبود بیشتر و تفکیک بهتر از قید تعامد بر روی بازنمایی داده ها استفاده می شود؛ همچنین اهمیت ساختار محلی نیز به حساب آمده است و در نهایت زیر مجموعه ای از ویژگی ها به کمک خروجی روش که ویژگی های امتیازبندی شده هستند انتخاب می شوند. برای ارزیابی عملکرد روش پیشنهادی، ازآنجاکه الگوریتم پرکاربرد در حوزه داده های بدون برچسب الگوریتم خوشه بندی است، زیر مجموعه ویژگی های به دست آمده در این الگوریتم مورداستفاده قرار می گیرند و با روش های متداول و مورد ارجاع در سایر کارها مقایسه می شوند