Browsing by Author "Kamel, Sherif"

Now showing 1 - 2 of 2

BREAST CANCER DETECTION USING DEEP LEARNING ON BIOMEDICAL MAMMOGRAM IMAGES
(Little Lion Scientific, 2024-04) Roy Chowdhury, Pretom; El-Dosuky, Mohamed; Kamel, Sherif
Millions of women worldwide are affected by breast cancer, which is a serious global health issue. The likelihood of successful therapy and the prognosis both greatly benefit from early identification. The most popular screening method for breast cancer, mammography, produces precise biological images that can help with the early detection of malignancies. However, it is still difficult to correctly interpret mammography pictures, which frequently results in false positives or negatives. This study attempts to create a biological mammogram based deep learning system for breast cancer diagnosis. Convolutional neural networks (CNNs) are used to automatically identify and analyse mammogram pictures in the proposed system, enabling radiologists to make quicker and more accurate diagnoses. To ensure the best performance during the training phase, these photos underwent preprocessing to reduce noise and enhance characteristics. The deep learning model used is a cutting-edge CNN architecture that was pretrained on a sizable dataset to fully utilise its learned representations. The deep learning model underwent thorough training, validation, and fine-tuning procedures to ensure robustness and generalizability. A variety of data augmentation methods, including rotation, scaling, and flipping, was used to enlarge and diversify the dataset during training. To further increase the model's accuracy, transfer learning was used to utilize knowledge from other similar tasks. Using a variety of criteria, such as sensitivity, specificity, accuracy, and F1 score, and the performance of the created breast cancer detection system was carefully assessed. The results showed a substantial increase in accuracy when compared to traditional mammography analysis methods. The method demonstrated impressive specificity in reducing false positives and sensitivity in identifying actual positive situations.
MACHINE LEARNING TECHNIQUES FOR CYBER SECURITY
(Little Lion Scientific, 2024-04) Sur, Soumik; El-Dosuk, Mohamed; Kamel, Sherif
Machine learning (ML) is a branch of artificial intelligence (AI) that focuses on developing computer programs that can recognize patterns in historical data, learn from it, and make logical judgements with little to no human input. Protecting digital systems, such as computers, servers, mobile devices, networks, and related data against hostile assaults is known as cyber security. Two key components of combining cyber security with ML are accounting for cyber security where machine learning is used and using machine learning to enable cyber security. This coming together may benefit us in a number of ways, including by enhancing the security of machine learning models, enhancing the effectiveness of cyber security techniques, and supporting the efficient detection of zero day attacks with minimal human interaction. The cyber security landscape has grown more complicated due to the quick development of technology, creating a number of difficulties for protecting sensitive data and important infrastructures. This project's objective is to implement three different systems using machine learning in cyber security. The first system investigates how reinforcement learning may be used to improve cyber security measures. Reinforcement learning algorithms are taught to make the best choices based on their interactions with the environment through trial and error, which can be useful in adjusting to changing cyberthreats. The second approach focuses on malware identification since evasive and polymorphic malware have proven difficult to identify using standard signature-based methods. Several machine learning and deep learning approaches are used in this effort to accurately identify and categorize dangerous software. The third solution uses machine learning and deep learning techniques to address the crucial problem of network intrusion detection. The performance of each system's machine learning models will be evaluated throughout the project using a variety of datasets alongside evaluation measures.