Bilgisayar Mühendisliği Bölümü Kitap Bölümü Koleksiyonu

Bu koleksiyon için kalıcı URI

Güncel Gönderiler

Listeleniyor 1 - 11 / 11
  • Öğe
    Prediction of software faults using machine learning algorithms and mitigating risks with feature selection
    (Elsevier, 2024) Bai, Femilda Josephin Joseph Shobana; Kaliraj S.; Ukrit, M. Ferni; Sivakumar V.
    Software fault prediction, a crucial component of software engineering, strives to detect probable flaws before they appear, thus enhancing the quality and reliability of software. Effective risk analysis is essential for reducing the risks and uncertainties that could arise during the development of software. The proposed work uses machine learning approaches to predict software faults and highlights the significance of risk analysis and feature selection. The accuracy of predictions can be increased by using feature selection approaches to help discover the features that strongly influence the prediction of software fault occurrence. The feature importance was identified by the algorithms using the decision trees (DT), gradient boosting machine (GBM), random forest (RF), and extreme gradient boosting (XGBoost) techniques. The models also underwent comparison by removing the features to understand the importance of the features and their correlation. Finally, a comparison is done to recognize the best model for software fault prediction.
  • Öğe
    Optimization of tree-based machine learning algorithms for improving the predictive accuracy of hepatitis C disease
    (Elsevier, 2024) Bai, Femilda Josephin Joseph Shobana; Jasmine, R. Anita
    Hepatitis C is a globally prevalent viral infection that has the potential to cause significant liver-related complications if not appropriately managed. The timely and precise identification of the medical condition is imperative for the efficient administration of patient care and therapy. One of the precise and potential diagnosis methods in the identification of hepatitis C is the utilization of machine learning (ML) algorithms. The present investigation focuses on the optimization of four ML algorithms which are tree-based algorithms, namely, random forest (RF), gradient boosting machines (GBMs), light gradient boosting machines (LGBMs), and extreme gradient boosting (XGBoost) with the aim of enhancing the predictive accuracy of hepatitis C disease. The investigation utilized a reliable dataset from the University of California, Irvine (UCI) Machine Learning Repository. The research methodology encompasses various stages, including data preprocessing, feature selection, hyperparameter tuning, and model evaluation. Optimization techniques, including the synthetic minority oversampling technique (SMOTE) for data balancing and grid search optimization for hyperparameter tuning, were utilized to improve the models’ performance. The optimized models were assessed through the utilization of stratified k-fold cross-validation and performance metrics, which comprise accuracy, precision, recall, F1-score, and area under the receiver operating characteristic (ROC) curve. The findings of our study indicate that the optimized tree-based algorithms exhibit superior performance compared to their nonoptimized counterparts. Specifically, LGBM demonstrated the highest level of predictive accuracy at 98.91%, followed by XGBoost at 98.70%, GBM at 97.83%, and RF at 97.29%. The LGBM learning approach has the potential to be broadly applied and extended to diverse medical datasets and use cases, thus advancing ML in the healthcare domain. The study highlights the importance of optimizing tree-based algorithms to improve the accuracy of early prediction of the prevalence of hepatitis C disease and promote patient health. This underscores the capacity of ML to improve healthcare outcomes. © 2024 Elsevier Inc. All rights reserved.
  • Öğe
    Neural networks
    (Elsevier, 2024) Pour Haji Kazem, Ali Asghar
    This chapter provides a comprehensive overview of neural networks as a powerful machine learning technique. It explores the fundamental concepts behind neural networks, shedding light on their structure and functioning. Beyond their technical aspects, the chapter delves into the significant role neural networks play in the realm of decision-making. By understanding how neural networks process and analyze data to make informed predictions and decisions, readers will gain valuable insights into the applications and implications of this technology in various domains. © 2024 Elsevier Inc. All rights reserved.
  • Öğe
    Reinforcement learning algorithms
    (Elsevier, 2024) Tareq, Wadhah Zeyad; Amasyalı, Mehmet Fatih
    The training scheme is a critical fundamental in multiagent systems, especially with reinforcement learning methods. The reinforcement learning agent builds its experience through interacting with the environment by trial and error. Later, the agent uses these experiences to decide which is the correct action and which one is not. In multiagent systems, two or more agents learn through trial and error in the same environment. These agents can cooperate to perform a single task or to compete to achieve a single goal. The training of multiple agents has many challenges. Selecting a suitable training scheme is one of these challenges. This chapter examines different schemes to find out the optimal scheme for training multiagent deep reinforcement learning. All applied schemes concentrated on two main fundamentals: centralized and distributed. All schemes tested on self-driving filed with multiple autonomous vehicles. Different traffic scenarios are utilized to measure the impact of each scheme in different situations. In the experiments, three different schemes were tested: centralized, distributed, and hybrid. The results show that the combined model (hybrid) achieves better performance compared with standard models.
  • Öğe
    Early detection of cardiovascular disease: Data visualization, feature selection, and machine learning algorithms for predictive diagnosis
    (Elsevier, 2024) Bai, Femilda Josephin Joseph Shobana; Ashok Kumar, Saranya; Maheswari M.; Aruna S. b; Krishnan, Aditya; Majid, Amaan
    Accurate and early diagnosis of cardiovascular disease is a big concern to improve the patient's well-being. The proposed research is focused toward the prediction of cardiovascular disease using a diversified dataset, which includes the patient's health history and diagnostic test results. The study focuses mainly on data visualization, feature selection, and predictive modeling. To identify the distribution of the features and the relationship between the features, data visualization was performed by using various plots and graphs. The important features in the dataset that can be helpful for better prediction are selected using embedded-based feature selection approach. The prediction of disease utilized machine learning (ML) techniques, including logistic regression (LR), decision trees (DTs), support vector machines (SVMs), and k-nearest neighbors (KNNs). F-1 score, precision, recall, accuracy, sensitivity, specificity, and receiver operating characteristic (ROC) curves are useful metrics for assessing how well machine learning models perform at predicting disease. These metrics provide insights into the advantages and drawbacks of the models, helping researchers to understand their effectiveness and suitability for specific tasks. 0.98, 0.82, 0.80, and 0.78 are the accuracies, and 0.99, 0.94, 0.89, and 0.83 are the area under the ROC curve (AUC) values of DT, KNN, SVM, and LR predictive models, respectively. The findings provide an insight to the healthcare professionals and researchers to understand the usefulness of predictive modeling for early predictions of cardiovascular disease. © 2024 Elsevier Inc. All rights reserved.
  • Öğe
    Mathematical programming
    (Elsevier, 2024) Hosseinzadeh Lotfi, Farhad; Saati, Saber; Shahriari, Mohammadreza; Rahmaniperchkolaei, Bijan; Taeeb, Zohreh
    Optimization stands as an exceptional cornerstone, harnessed extensively and with a resounding impact to confront the multifaceted challenges of the tangible world. Its expansive repertoire of applications, complemented by its innate ability to unfurl optimal solutions, has bestowed upon it an inherent allure, captivating a diverse spectrum of enthusiasts traversing the intricate landscape of research. Spanning domains as varied as the fundamental sciences, engineering marvels, strategic management paradigms, agricultural landscapes, and the ecological tapestry of environmental studies, optimization resonates as a unifying force in problem-solving endeavors. Nestled within the confines of this chapter is a captivating odyssey, a journey that navigates through the bedrock of optimization's essence. We embark upon this voyage by peeling back the layers of fundamental concepts that lay the groundwork for this dynamic field. From this foundational standpoint, we plunge headfirst into the realm of optimization problems, where two distinctive categories beckon: the structured world of linear programming and the intricate enigma of nonlinear programming. As we unravel the mysteries that enshroud these two archetypes, we simultaneously unveil the symphony of solution algorithms meticulously tailored to navigate each unique category's labyrinthine complexities. In our pursuit to instill a comprehensive comprehension, we seamlessly weave tangible examples throughout the narrative tapestry. These instances of real-world conundrums serve as eloquent companions to the theoretical frameworks, acting as guiding beacons to illuminate the path for readers. Through this harmonious marriage of theory and application, we aspire to bestow upon our readers the twin gifts of knowledge and proficiency—a twin that empowers them to not only navigate the intricate terrain of optimization but also to rise as adept problem solvers, armed with the analytical acumen and innovative strategies requisite to surmount the most formidable challenges that reality presents. © 2024 Elsevier Inc. All rights reserved.
  • Öğe
    Deep learning
    (Elsevier, 2024) Tareq, Wadhah Zeyad Tareq
    Extracting features from raw data such as videos and soundtracks is difficult and almost impossible. Handcrafted design and feature engineering are hard tasks and need experimentation, evaluation, and creativity. Feature engineering requires exploring the data and the impact of each feature before selecting them. This exploration process requires significant time and effort. Deep learning is a powerful approach to extracting features by transforming raw data into numerical features. This process occurs by using nodes known as neural networks such as the human brain. Neural networks contain multiple layers, and each layer can extract several features from the input data. In this chapter, the convolutional neural network is used to extract features from video games. These features describe the state or observation of the game to enable an intelligent agent to make decisions. The deep learning algorithm combined with the reinforcement learning approach to build a deep reinforcement learning agent able to play different games using game screenshots and game scores just like a human player. The results showed that the agent is able to enhance its performance after several steps, which proves the efficiency of feature extraction using deep learning algorithms. © 2024 Elsevier Inc. All rights reserved.
  • Öğe
    Classification and clustering
    (Elsevier, 2024) Tareq, Wadhah Zeyad Tareq; Davud, Muhammed
    Data analysis is the process of understanding or extracting patterns from raw data. One of the widely used methods in data analysis is machine learning. Machine learning is a system or model that can learn from raw data to make decisions without human intervention. Classification and clustering are the most popular machine learning technologies for the analysis of data. Each technology involves many algorithms that aim to categorize objects into classes depending on the object's features. In this chapter, we introduce a guide to both classification and clustering technology by applying different algorithms to different datasets. The classification dataset differs from the clustering dataset. The reason here is to explain the suitable type of data for each technology. For clustering, the k-means clustering algorithm is applied. For classification, the decision tree algorithm is applied. The results showed the efficiency of different machine learning algorithms for data analysis and decision-making. © 2024 Elsevier Inc. All rights reserved.
  • Öğe
    (Artificial) neural networks
    (Elsevier, 2024) Tareq, Wadhah Zeyad Tareq
    Buying a new or used car is one of the challenges faced by people in most parts of the world. A car's price can be affected by many factors such as economic and political crises. With the high demand for cars and the different crises across the world, predicting car prices has become impossible and incompetent. This study aims to find the overall amount that customers can offer to purchase a car. The suggested model uses artificial neural networks (ANN) to predict customer budgets based on different features. Moreover, this chapter provides a guideline for the authors and researchers to understand ANN's basic concepts and ideas to be used later in different areas. The suggested model was trained using 500 samples of customers from different countries, with different ages and different annual salaries. The proposed model obtains the highest accuracy of 95%. © 2024 Elsevier Inc. All rights reserved.
  • Öğe
    Reinforcement learning and trustworthy autonomy
    (Springer International Publishing, 2018) Luo J.; Green S.; Feghali P.; Legrady G.; Koç, Çetin Kaya
    Cyber-Physical Systems (CPS) possess physical and software interdependence and are typically designed by teams of mechanical, electrical, and software engineers. The interdisciplinary nature of CPS makes them difficult to design with safety guarantees. When autonomy is incorporated, design complexity and, especially, the difficulty of providing safety assurances are increased. Visionbased reinforcement learning is an increasingly popular family of machine learning algorithms that may be used to provide autonomy for CPS. Understanding how visual stimuli trigger various actions is critical for trustworthy autonomy. In this chapter we introduce reinforcement learning in the context of Microsoft's AirSim drone simulator. Specifically, we guide the reader through the necessary steps for creating a drone simulation environment suitable for experimenting with visionbased reinforcement learning. We also explore how existing vision-oriented deep learning analysis methods may be applied toward safety verification in vision-based reinforcement learning applications. © Springer Nature Switzerland AG 2018.
  • Öğe
    Mathematical optimizations for deep learning
    (Springer International Publishing, 2018) Green, Sam; Vineyard, Craig M.; Koç, Çetin Kaya
    Deep neural networks are often computationally expensive, during both the training stage and inference stage. Training is always expensive, because back-propagation requires high-precision floating-pointmultiplication and addition. However, various mathematical optimizations may be employed to reduce the computational cost of inference. Optimized inference is important for reducing power consumption and latency and for increasing throughput. This chapter introduces the central approaches for optimizing deep neural network inference: pruning "unnecessary" weights, quantizing weights and inputs, sharing weights between layer units, compressing weights before transferring from main memory, distilling large high-performance models into smaller models, and decomposing convolutional filters to reduce multiply and accumulate operations. In this chapter, using a unified notation, we provide a mathematical and algorithmic description of the aforementioned deep neural network inference optimization methods. © Springer Nature Switzerland AG 2018.