MPhil Thesis

Browse

Search Results

Now showing 1 - 10 of 220
  • No Thumbnail Available
    Item
    An End-To-End Efficient License Plate Detection and Recognition System using Deep Learning
    (Daffodil International University, 2025-01-13) Bristi, Nushrat Jahan
    This research presents an enhanced license plate recognition system for real-time detection and recognition in transportation and security applications. YOLO object detection algorithms (YOLOv8s, YOLOv8x, YOLOv11s) enable accurate license plate localization, while EasyOCR ensures reliable alphanumeric identification in challenging situations, including low light and complex backgrounds. Testing on diverse datasets demonstrated high accuracy, with YOLOv11 and data augmentation achieving a peak F1 score of 98%. The system also addresses Bengali character recognition challenges, offering a foundation for region-specific improvements. These outcomes validate the system's effectiveness for law enforcement, traffic management and security.
  • No Thumbnail Available
    Item
    A Framework for Human Skin Disease Classification Using Convolutional Neural Network
    (Daffodil International University, 2025-01-20) Hera, Mst. Dilruba Yeasmin
    One of the most dangerous types of cancer is skin cancer, it becomes a significant health hazard when not treated and detected on time. Skin cancer may spread to other parts of the body and complicate treatment if it is not detected in its early stages. Mainly it is the result of abnormal skin cell growth, usually the cells are stimulated by the sun for a long time. The early detection of skin tumors is a basic but highly complicated and expensive process due to the complexity of the diagnostic methods implicated. The identification of skin cancer by the location and cells involved augments the necessity of a very precise classifier for a successful diagnosis. Where the use of CNN in the recognition and classification of skin cancer, especially in skin lesion classification has been proposed to solve this issue. The utilized diagnosing method includes the utilization of image processing algorithms and deep learning models to increase accuracy and efficiency. Methods like image augmentation are then used for adding more rows to the dataset are used to scale up the dataset. This way, the model understands the diverse cases encountered. In addition, transfer learning is useful for increasing the classification accuracy by using pre-trained models for improved performance. As one of deep learning's deep architectures, CNNs serves as a key player in the extraction of features and in the classification of skin problems like psoriasis. This technique has been impressively productive for it gets a hit rate of 75%, thus revealing future prospects in the medical field.
  • No Thumbnail Available
    Item
    Improved Explainable Educational Data Mining System for Enhancing Programming Skills
    (DAFFODIL INTERNATIONAL UNIVERSITY, 2024-07-18) Mohamud, Mohamed Abdulle
    Forecasting student academic performance benefits from the extremely effective method known as educational data mining (EDM), which also helps to find important links within educational data. Evaluating and improving students' programming competency has been the main emphasis of many recent studies. Still, there are chances for constant development in this field. In this work, we provide an improved and understandable Educational Data Mining (EDM) approach for spotting and improving students' programming capacity. This proposed EDM system seeks to investigate a very effective feature engineering approach, a suitable classification technique, and the use of Explainable Artificial Intelligence (XAI) tools for model explanation. We do ablation study to find the best feature engineering method. The categorizing process decides students' current programming state. Six basic Machine Learning (ML) algorithms—decision tree, Support Vector Machine, Random Forest (RF), artificial neural network, Naive Bayes Classifier, k-Nearest Neighbor, and Ensemble method—are the main subjects of this module. Many criteria—including accuracy, precision, recall, f1-score, ROC curve, McNamar test, and others—are used to assess the performance of these algorithms. The experimental results show that among the many models, the Random Forest (RF) and the Stacking-SRDA ensemble technique can classify the students with more accuracy than others. To improve the interpretability of the model, we have finally used XAI technologies like Eli5, SHAPASH, and Local Interpretable Model Agnostic Explanations
  • No Thumbnail Available
    Item
    Analysis of Reusability of Used Clothes Using Machine Learning Algorithms
    (DAFFODIL INTERNATIONAL UNIVERSITY, 2024-01-28) Rubel, Md. Salauddin
    With $45 billion in apparel exports in 2022, Bangladesh is the second-ranked country in the world for apparel exports also the second largest producer of textile waste. The Bangladeshi apparel sector is expected to generate US$10.15 billion in revenue by 2023. Purchasing power of Bangladeshi people have increased and expenses on apparel product has also increased. The habit of repeating clothes that worn once has decreased which makes our wardrobe filled with lots of rarely used clothes and after a certain time we throw them as a wastage. Our study is aimed to develop a machine learning algorithm to predict clothing reusability. For our model we use clothing type, fabric type, usage duration, damage, distortion and color information of a used cloths. We use Classification algorithms for constructing our predictive model. We have applied five classification machine learning algorithm which are Decision Tree, Random Forest, Naïve Bayes, Logistic Regression and SVM. With the given information of a used cloth our model can predict the reusability option for it, the options are: resale, reuse and turn into jhoot product. By reselling a used cloths one can earn save some money and on the other hand people having less money can get a good product. Reusing clothing items means using to create a new apparel item or using in home craft. The last option of reusing is turning into jhoot products, cloths which have used more than their average life cycle are used in jhoot. This research achieved model accuracy between 79% to 85% on predicting reusability of different apparel items. Future study will explore new machine learning approaches with larger dataset and also enable a system that will be useful for textile industry to achieve sustainability in clothing wastage.
  • No Thumbnail Available
    Item
    Forecasting and Comparison of Economic Indicators for The Universal Pension Scheme in Bangladesh
    (DAFFODIL INTERNATIONAL UNIVERSITY, 2024-01-28) Hasan, Md. Warid
    Article 15 of the Bangladeshi constitution requires the state to guarantee "the right to social security." From birth to death, the Universal Pension Scheme aims to provide social security for all of its citizens and ensure that no one is left behind. This universal pension scheme is expected to provide sustainable and well-structured social security to the population of Bangladesh especially the growing elderly population due to the increase in average life expectancy. Bangladeshi citizens of all classes and professions can participate in this pension scheme between the ages of 18 to 50 years as per the national identity card or conditionally by making a minimum contribution of 10 years. The government is announcing a total of six pension schemes, but for now, four have been launched, namely PROBASH, PROGOTI, SHUROKKHA and SAMATA. Currently 62% of our total population is working. Although the current average life expectancy is 72.3 years, there is a possibility of further increase in the future, the increase in the average life expectancy and the increase in the number of single households will increase the dependency ratio in the future, so it is necessary to build a sustainable social security structure. As much as this pension scheme will benefit the people of the country, it can also harm them terribly if the government does not implement this system in a proper way. So, we will discuss a system that will most likely verify the feasibility of our pension system, as well as how much it will be acceptable to people and the possible transparency of the pension scheme. We can highlight some studies that can give us a better idea of the past and future of our economy. Among these, past data visualisation is a very significant method for reviewing the past. And I think data forecasting is the most appropriate decision for the future. We will do a good comparison of our own economic data with the data of other countries so that we can better understand the feasibility of introducing our pension system. And we will use the forecasting model so that we can see the possibility of the future of the data. In this way, we can get an idea of how the future development of the country's economy can affect our pension system. Our studies will surely help the financial sector of the government and also give a clear and proper understanding of this pension scheme among the people of the country.
  • No Thumbnail Available
    Item
    Rice Leaf Disease Detection Using Machine Learning Technique
    (Daffodil International University, 2024-07-13) Hasan, Md Mehedi
    This study explores the application of deep learning models for the detection of rice leaf diseases, a critical issue impacting global rice production and food security. The research focuses on five advanced deep learning architectures: Convolutional Neural Network (CNN), Xception, VGG19, MobileNetV2, and InceptionResNetV2. Utilizing a dataset comprising 6,420 images across four disease categories—Brown Spot, Tungro, Bacterial Blight, and Blast—each model was trained and evaluated to determine its accuracy and effectiveness in disease classification. The proposed methodology encompasses data collection, labeling, image processing, model selection, training, evaluation, and testing. Results demonstrated that the CNN model achieved the highest accuracy at 98.44%, followed closely by MobileNetV2 at 97.82%, VGG19 at 96.57%, InceptionResNetV2 at 95.43%, and Xception at 95.07%. These high accuracies underscore the potential of deep learning models in early disease detection, which is crucial for timely intervention and effective crop management. Comparative analysis with traditional machine learning approaches such as Support Vector Machines (SVM) and Decision Trees, which typically yielded lower accuracies between 81.8% and 97%, highlights the superior performance of deep learning techniques. Furthermore, the study discusses the ethical considerations, including data privacy, accessibility for small-scale farmers, and the need for unbiased models, ensuring equitable benefits across diverse agricultural contexts.
  • No Thumbnail Available
    Item
    Deepretina: Deep Learning Approach To Detect Retinal Abnormality In Computer Vision
    (DAFFODIL INTERNATIONAL UNIVERSITY, 2024-01-30) Nasrin, Sonia
    Currently, almost 1.2 million people in our country are blind, while around 3.51 lakh people have low vision. The pattern of eye abnormalities is changing along with an increasing rate of dry eye, cornea-related problems and eye problems related to diabetes. Early identification of eye diseases especially retinal abnormality plays a vital role to prevent the blurry vision in patients. In my research, a hybrid deep learning model is proposed to detect retinal abnormality by scanning a single retinal image of a patient. First, a new multi-label retinal disease dataset, Retinal Fundus Multi-Disease Image Dataset (RFMiD) version 02 is collected from a renewed journal website “Multidisciplinary Digital Publishing Institute” (mdpi), where 46 retinal diseases labels are available with high resolution. Next, dataset is going through analysis and preprocessing techniques to deals with data imbalance and large size (8gb) problem. Numerous analysis and experiments are performed to evaluate the models for better results. In the model, Convolutional neural models – EfficientNet, VGG16, NesNetMobile are used to analysis comparative result as well. EffectiveNet gives the highest accuracy among them and that is 85%. Voting Ensemble method is used to increase model accuracy (88%) for better prediction than could be gained from any of the constituent learning algorithms. This model is used to detect normal or abnormal retinal conditions for early treatment.
  • No Thumbnail Available
    Item
    A Study to Analyze The Reviews of E-commerce Business in Bangladesh Using Machine Learning Techniques
    (DAFFODIL INTERNATIONAL UNIVERSITY, 2024-02-07) Naim, Md. Jannat-Ul
    In the fast-changing landscape of e-commerce, establishing trust has become essential for sustainable development, particularly in regions like Bangladesh where customer confidence is poor. This study analyzes customer reviews in the Bangladeshi e-commerce industry using few of the Machine Learning approaches, especially utilizing Natural Language Processing (NLP). The primary focus is on identifying patterns and sentiments in Bengali reviews in order to educate businesses about consumer preferences as well as enhance services, ultimately building trust. This research work evaluates effectiveness of very well-known Machine Learning technique call natural language processing for processing a huge number of Bengali e-commerce reviews while considering specific linguistic and contextual differences. The study also intends to derive useful insights from consumer feedback analysis in order to improve products and services, enhance customer satisfaction, and achieve a competitive edge in the Bangladeshi e-commerce environment. Additionally, the analysis of sentiment findings is expected to be beneficial to the growth and development of Bangladesh's e-commerce sector, with an eye on possible challenges with implementing data-driven initiatives that depend on consumer feedback. The research methodology is based on collecting data from Kaggle and efficiently pre-processing text using Natural Language Processing. To gain more precise outcomes of my models, unigram, bi-gram, and trigram features will be integrated into the Linear Support Vector Machine, Karnel Support Vector Machine, Random Forest Classifier, Decision Tree Classifier, Naive Bayes Classifier and Logistic Regression techniques. The research study has achieved 90.76% accuracy using the Random Forest algorithm utilizing Bigram feature.
  • No Thumbnail Available
    Item
    Customer Satisfaction Sentiment Analysis For Online Transactions In Bangladesh.
    (DAFFODIL INTERNATIONAL UNIVERSITY, 2024-07-10) Pias, Md. Almahmud
    Online transactions are becoming increasingly popular in Bangladesh, starting from shopping, we transact online. Online payment plays a very important role in paying any shopping bill. To find any possible problems and enhance the user experience as a whole, it is crucial to comprehend the degree of client satisfaction with these kinds of transactions. With an emphasis on aspects including platform usability, transaction security, and overall customer experience, the poll sought to determine Bangladesh's degree of customer satisfaction with online transactions. A report based on a sample of customers through which we understood how customers are satisfied with online payment and got an idea to understand their opinion. survey's findings indicate that Bangladesh's mean consumer satisfaction rating for online transactions is 6.2096813773 out of 10. Through this we came to know that all the customers of Bangladesh are satisfied with online shopping and everyone can improve from everyone's place. Through this test, we can know more about the value and satisfaction of online transactions in Bangladesh. They have largely responded to improving customer experience and digital transactions. All in all, a satisfactory established informed that online shopping plays an important role.
  • No Thumbnail Available
    Item
    Impact of Artificial Intelligence on Tertiary Level Education in Bangladesh: A Comparative Analysis of Positive and Negative
    (DAFFODIL INTERNATIONAL UNIVERSITY, 2024-07-11) Dowlla, Md. Asaf-Ud
    This thesis explores the integration of Artificial Intelligence (AI) in tertiary-level education in Bangladesh, examining the perceptions, usage, and impact of AI among students from various universities. A survey conducted with 787 respondents from private, public, national, and other tertiary institutions revealed that 70.6% of students are familiar with AI in education, with 73.4% having used AI-powered tools. A significant majority (69.1%) believe AI can be beneficial for tertiary education, and 71.8% feel that AI enhances their learning experience. However, there is a notable degree of skepticism, with 51.3% feeling guilty about using AI for assignments, and 72.5% believing stricter regulations are necessary. Comparative analysis with global data highlights that Bangladeshi students are more familiar and optimistic about AI's benefits compared to their global counterparts. While 73.5% of Bangladeshi students reported improved grades due to AI, 63.4% believe it makes the education system more effective and efficient. The study also reveals significant gaps in satisfaction with current technology integration, with only 15.7% expressing satisfaction, indicating a need for further development in this area. The findings underscore the potential of AI to transform tertiary education in Bangladesh, while also pointing to areas for improvement and the need for balanced regulations. Future research should focus on increasing the survey population to include a more diverse range of institutions and disciplines, and consider longitudinal studies to track the evolving impact of AI in education