Project Report

Browse

Subject: search.filters.subject.Machine learning

Search Results

Now showing 1 - 10 of 251
  • Thumbnail Image
    Item
    Enhancing Cloud Security Using Artificial Intelligence: Analyzing the Effectiveness and Challenges of AI-Powered Threat Detection and Prevention Readiness
    (Daffodil International University, 2025-09-20) Adnan, MD. Sanaul Islam
    The rapid development of cloud computing has changed the way organizations store, retrieve, and secure data, providing incredible scalability at a cost and posing new and challenging cyber threats that most security tools are incapable of determining effectively. In this thesis, artificial intelligence is explored to enhance the threat detection and prevention preparedness in the cloud setting, using a well-designed hybrid dataset that combines the data on synthetic network traffic with more detailed information on the governance of different organizations. Several models of AI were also evaluated, such as Random Forests, XGBoost on the basis of detecting various types of attacks, as well as the models of Logistic Regression, Gradient-Boosted Trees, and Multilayer Perceptions on the basis of predicting the security posture maturity. The results indicate that AI can provide and identify cyber threats with very high precision and issue corresponding ratings of organizational readiness to prevent an attack. It is observed that the XGBoost had the best accuracy of detection and Logistic Regression models provided the interpretable and accurate prediction of governance score. This two-pipe approach to reactive detection and proactive prevention is a response to the comprehensive needs of the modern cloud security. However, there are a number of difficulties, in particular, to achieve the transparency of the models, protect the sensitive information, and be ready to adjust to the new changes in the nature of threats in the multifaceted clouds. Furthermore, the factors of such consideration of the operation, as the rapidity of the detection, as well as the continuity between the new security setups and the old ones, are also paramount. The thesis mentions such directions of future research as the development of adaptive AI systems that would enable it to learn in real-time, collaboratively privacy-safe methods, and more fundamental integration with zero-trust systems and the development of explainable AI tools as the tools of building trust and ensuring effective human oversight. Overall, this paper has very good information that the use of intelligent artificial intelligence to secure the cloud can be valuable in a balanced way to false threat detection methods and efficient governance check-ups as a good foundation of organizations that plan to deploy intelligent, responsible, and trustful defenses on the cloud.
  • Thumbnail Image
    Item
    An Intelligent Technique for Thyroid Disease Detection Using Machine Learning
    (Daffodil International University, 2024-07-24) Bin Hafiz, Md. Shahnewaj
    An underactive thyroid gland is the hallmark of the common endocrine condition hypothyroidism, which can cause a variety of health problems. A timely and precise diagnosis is essential for the proper management and treatment of this illness. In this paper, we investigate how machine learning approaches—more especially, ensemble techniques like Bagging and Boosting—can be used to forecast hypothyroidism. We have taken two popular datasets from Kaggle and Figshare website. We use a wide range of data, such as laboratory and clinical characteristics, to train and assess various machine learning models. The Bagging method lowers variance and improves overall model stability by combining predictions from several base learners. By giving misclassified cases a larger weight, the technique known as "boosting" aims to repeatedly improve the model's accuracy. The most accurate classifier was the traditional technique, which achieved an impressive accuracy rate of 93.17% by Random Forest (RF). Other classifiers that were used included Logistic Gradient Boosting (GB), Regression (LR), Adaboost Classifier (ABC), K-Nearest Classifier (KN), Support Vector Machine (SVM), Decision Tree (DT), Ridge Classifier (RC), Quadratic Discriminant Analysis (QDA), Passive Aggressive (PA), Gaussian Naïve Bayes (GNB). In addition, 92.15% accuracy was obtained by the Boosting Gradient Boosting (GB), while Boosting Random Forest (RF) 91.86% accuracy was attained. Hyperparameter tweaking was used to maximize each classifier's performance. After conducting an experimental examination and reviewing prior research, it was determined that the Random Forest (RF) classifier performed very well, correctly diagnosing hypothyroid illness with an astounding accuracy rate of 93.17%.
  • Thumbnail Image
    Item
    Lung Cancer Classification Using Machine Learning Techniques
    (Daffodil International University, 2024-07-24) Sen, Ankan
    In this paper, I explored the task of categorizing medical images into three categories: benign, malignant, and normal. The dataset comprises images which are preprocessed and resized for model training. After preprocessing, the dataset consists of 877 training samples and 220 testing samples. Each image is represented as a grayscale image with a single channel. The labels for the images are provided as one-hot encoded vectors, with each label indicating the class of the corresponding image. The medical images used in this study are sourced from a curated dataset specifically designed for research in medical imaging analysis. This dataset contains a diverse range of images capturing various medical conditions, allowing for comprehensive training and evaluation of the classification model. The primary objective of this research is to develop a classification model capable of accurately distinguishing between the three classes of medical images. To achieve this, I plan to employ convolutional neural network (CNN) architectures, which have demonstrated strong performance in image classification tasks. By leveraging CNNs, I aim to capture relevant features from the medical images and utilize them for effective classification. The evaluation of the classification model will be conducted using the testing dataset, where the model's performance will be assessed based on metrics such as accuracy precision, recall, and F1-score. Additionally, the model's generalization capability will be analyzed to ensure its effectiveness in classifying unseen data. The outcome of this research holds significant implications for medical diagnostics and healthcare applications. Accurate classification of medical images can aid healthcare professionals in identifying and diagnosing various medical conditions, potentially leading to timely interventions and improved patient outcomes. Furthermore, the developed classification model can serve as a valuable tool for automated image analysis, augmenting the capabilities of medical practitioners and enhancing the efficiency of diagnostic processes.
  • Thumbnail Image
    Item
    Ehancing cotton leaf disease detection and classification through machine learning and deep learning techniques
    (Daffodil International University, 2024-07-24) Alim, Md. Abdul
    In Bangladesh, cotton has the potential to be a significant revenue crop. To accommodaterising demand, we import 3 billion dollars of cotton yearly (The Business Standard). There isn't an alternative to this issue but to grow cotton. However, the most prevalent issue among farmers was diagnosing crop illness by applying the antiquated growidea. They are unable to identify crop diseases early enough to treat the crops withtheappropriate measures. Particularly in rural areas where farmers suffer fromimproper knowledge leading to crop disease identification. The study demonstrates manyalgorithms and deep learning methods for cotton leaf disease detection. I utilize anMLframework that includes three deep-learning models in it.Model accuracy are compared, and the results show that different architectures perform differently on the task. Compared to the other models, CNN's accuracy is somewhat lower at 82.35%. Theaccuracy of VGG16 is greatly improved to 97.69%, demonstrating its usefulness inthis situation. ResNet50 does well too, with 92.51% accuracy. With the highest accuracyof 99.28%, XceptionV3 beats all other models, proving its exceptional ability to completethis assignment. InceptionV3, which has an accuracy of 94.42%, likewise performs admirably. In conclusion, XceptionV3 has the best accuracy at 99.28%, while CNNhas the lowest accuracy at 82.35%.
  • Thumbnail Image
    Item
    A Machine Learning Approach For Dengue Disease Prediction
    (Daffodil International University, 2024-07-24) Haque, Md. Majedul
    The fact that dengue fever is a global health threat, that millions of cases are recordedevery year, is a matter of urgency. The prompt and accurate identification of this virus, which is transmitted by mosquitoes, is a prerequisite for treatment and control. Theexperts predict 400 million cases and 25,000 deaths in the process. It is a major publichealth problem a year. Machine learning algorithms, which have recently proven tobeaviable and promising tool in medical diagnostic, can be a possible method of diagnosingdengue fever. In this work, a machine learning-based method of diagnosing dengue fever using the numerical data given by the patients is brought to life. Dengue fever, amosquito-borne illness, is a tropical and subtropical disease that is native to these regions. The first step is the diagnosis and treatment of the disease as soon as possible in order toprevent the occurrence of the major sequelae and the death caused by dengue fever. Machine learning and artificial intelligence are applied for the purpose of datainterpretation and prediction. It has been verified as a successful tool in the detectionof dengue fever from a broad range of data sources, including blood samples, medical records, and environmental data. This study provides a machine learning solutionfor dengue fever diagnosis. Our method is based on a fusion of various factors that we import from clinical records and blood samples to design a classifier. We evaluated our randomforest method on a dataset made up of 820 patients and we proved that it can reach 98%accuracy. This paper shows the possible use of machine learning in the early diagnosis of dengue infection. The improvement of the diagnosis and treatment of dengue fever, this may eventually lead to a decrease in the death rate of this disease. The current methods of diagnosing dengue fever are mainly expensive, slow and not always accurate. Thanks tothe ML of the diagnostic systems for dengue fever, we can now create themthat will bemore accurate and efficient. Machine learning algorithms can be taught using the hugeamount of clinical and image data which are related to dengue fever, and thus, theycanbe used to identify the patterns of dengue fever.
  • Thumbnail Image
    Item
    AI-Driven Bone Fracture Detection:
    (Daffodil International University, 2024-07-24) Emon, Mir Md. Asif Jahan; Tisha, Esrat Jahan
    This study, titled "AI-Driven Bone Fracture Detection: Leveraging Image Processing and Machine Learning on X-ray Images," embarks on enhancing the accuracy and efficiency of diagnosing bone fractures using advanced AI techniques. Utilizing a dataset of X-ray images augmented with metadata on patient demographics and clinical details, several deep learning models, including VGG16, MobileNetV2, InceptionV3, ResNet50, and hybrid combinations, were trained and validated. These models demonstrate substantial promise in identifying and classifying bone fractures with varying degrees of precision. This study gets a high accuracy of 89% in MobileNetV2 while using fully raw data. The research highlights the superior performance of MobileNetV2 and hybrid models, which combine the strengths of multiple neural network architectures to optimize fracture detection. By integrating these AI models into clinical settings, the study aims to alleviate the workload on radiologists, expedite diagnostic processes, and potentially enhance patient care by offering rapid and accurate fracture evaluations. Moreover, the study explores the ethical dimensions of AI deployment in medical diagnostics, focusing on data privacy, bias mitigation, and system transparency. As the integration of AI in healthcare progresses, this research paves the way for future explorations into expanding the models' capabilities to other medical imaging modalities and developing real-time diagnostic tools. This work not only advances the field of medical AI but also sets a benchmark for future research aimed at refining AI-driven diagnostics in healthcare.
  • Thumbnail Image
    Item
    A Comparative Study of Machine Learning Algorithms for Property Price Forecasting
    (Daffodil International University, 2024-07-24) Fahim, Dehan Arif; Nayem, Irfan Ahamed
    As the housing market grows, it's important to estimate pricing for both businesses and individuals. Nonetheless, a number of variables influence changes in home prices. Bangladesh is an overpopulated country, therefore a number of interrelated factors affect the price at which real estate is sold.The property's amenities, location, and size are crucial factors that could affect the cost.The objective of this investigation is to predict the prices of property in Bangladesh by employing a variety of machine learning algorithms. In order to guarantee precise predictions and robust model training, we assembled a comprehensive dataset of 18,835 property listings from Bproperty & Bikroy.com. Random Forest, Support Vector Machine (SVR), Decision Tree, XGBoost, CatBoost, and LightGBM are among the algorithms implemented in this investigation. The performance of these models was assessed using a variety of metrics, including R-squared, Mean Absolute Error (MAE), Root Mean Absolute Error (RMAC), and Mean Squared Error (MSE). The CatBoost and XGBoost models achieved the highest R-squared value of 91%, indicating superior accuracy, but XGBoost performed slightly better with less RMSE, MAE, MSE value. While the DecisionTrees model yielded the lowest R-squared value of 82%, indicating relatively poorer performance. The value of our findings and the potential for future research in property market analytics are underscored by the effectiveness of advanced machine learning models, particularly CatBoost, in predicting property prices within the Bangladeshi market. This information is valuable for stakeholders.
  • Thumbnail Image
    Item
    Soil Condition Monitoring for organic tea plantation by using machine learning
    (Daffodil International University, 2024-07-24) Hossain, Md. Emran; Dristy, Dil Afrose
    In tea farming, it is important to monitor soil conditions so as to maintain good health of the soil and a high-quality tea. In this research, machine learning techniques are employed to optimize soils management practices in these plantations. For instance, various parameters including pH, potassium, calcium and magnesium contents were tested on loamy soils obtained from different areas. Data has been pre-processed and used for training different machine models like Random Forest, Gaussian Naïve Bayes, Decision Tree, Support Vector Classifier (SVC), Multi-Layer Perceptron (MLP), among others. Among the above-mentioned ones with a highest accuracy rate of 99.18% Random Forest was able to predict accurately the soil condition having effect on management of the same. This finding will be important to farmers and agricultural scientists who would like to enhance organic farming with the help of data analysis. In this study, these models had a very high predictive capability when it came to soil status. It indicated that Random Forest was the most accurate with 99.18% accuracy implying that it is good at analyzing soil data and providing reliable predictions for the same. These were closely followed by Gaussian Naive Bayes, Decision Tree, SVC and MLP which were other models that also performed well with accuracies of 98.36%, 98.63%, 98.77% and 97.95% respectively.
  • Thumbnail Image
    Item
    Dengue Disease Prediction by Machine Learning Algorithm in Bangladesh
    (Daffodil International University, 2024-07-24) Kibria, Md. Golam; Ha-mim, Fatema
    In Bangladesh, dengue fever is still a major public health issue that presents difficult management and preventative strategies. Proactive steps to lessen dengue's negative effects on public health can be made easier with accurate dengue epidemic prediction. In this work, we examine how well different machine learning methods predict the incidence of dengue fever in Bangladesh. We create a dataset with 1000 observations overall that consists of 9 input attributes and 1 outcome attribute. The input attributes comprise clinical markers such as NS1, IgG, and IgM levels in addition to demographic data like age and gender. Our prediction algorithm also incorporates geospatial variables like district, size, and kind of place. The output property "Outcome" indicates if dengue fever has occurred; this is a binary classification operation. The predictive performance of six machine learning algorithms is assessed, including Logistic Regression, Random Forest, Decision Trees, AdaBoost, Extreme Gradient Boosting (XGBoost), and LightGBM. With an astounding 98.67% accuracy rate, Random Forest is the most accurate of these algorithms. Our results highlight the potential of machine learning methods, especially Random Forest, to accurately forecast the incidence of dengue illness in Bangladesh. With the use of these prediction models, dengue outbreaks may be detected early and managed proactively, allowing for the prompt deployment of resources and the execution of focused intervention techniques to lessen the disease's crippling effect on public health systems.
  • Thumbnail Image
    Item
    Soccer player’s suitable playing position prediction using machine learning
    (Daffodil International University, 2024-07-24) Patoary, Md. Arman Hosen; Shil, Rudra Prashad
    Accurately identifying the most suitable position for a football player is crucial for optimizing team performance and player development. This paper will investigate the potential of machine learning algorithms in predicting player positions based on various performance metrics. We propose a novel approach that will utilize Logistic Regression, K-Nearest Neighbors, Multi-Layer Perceptron, Random Forest Classifier, Support Vector Machine classifier, and Decision Tree to analyze a comprehensive dataset of player attributes, including physical stats, skill assessments, and positional data. The model will be evaluated on its ability to correctly predict the primary positions of players across different leagues and levels of competition. The results will demonstrate that our proposed approach achieves high accuracy which is in Logistic Regression and that is 75% Accuracy, in predicting a player's suitable playing position. Furthermore, we analyze the feature importance scores to gain insights into the key attributes that are most influential in determining player positions