Journal of Al-Qadisiyah for Computer Science and Mathematics

Skin lesion detection is a very important task in medical applications such as disease diagnostic and cancer detection. This paper suggests an efficient algorithm for skin lesion detection relying on advanced image processing techniques without using machine learning or deep learning algorithms. A set of algorithms is proposed to enhance traditional image processing methods to scrutinize dermatological images, concentrating on melanoma detection. The proposed method combines color space transformation, contrast enhancement, active contour segmentation. ABCD criteria (Asymmetry, Border irregularity, Color variegation, and Diameter) is used in feature extraction. The approach is evaluated on the ISIC (International Skin Image Collaboration) 2020 dataset, accomplishing an accuracy of 84% in differentiating between malignant and benign skin lesions. The results indicate the potential of non-machine learning techniques in dermatological image analysis, providing computationally efficient and interpretable alternative to data-intensive deep learning algorithms.

In this paper, we compare our method on DEFRAUD with current online API services and the state-of-the-art machine learning model for defending against phishing websites. Rule-based methods, in nature traditional such rules tend to get outdated quickly and are not capable of tracing new tactics used by the malicious ware every second. Over time, new proactive approaches enabled by machine learning (ML) have become more important as these solutions are flexible and adaptable in their ability to scan through modern data breaches for patterns from millions of datasets. In this study, we explore various machine learning algorithms: Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Trees (DT) Random Forest (RF), Support Vector Classifiers (SVC) and xgBoost for phishing website detection. Ensemble Methods like Random Forest, XGBoost have better accuracy/precision/recall. Metrics. While XGBoost is resource hungry, it is well known for out of the box support with huge data dimensions as well deep learning framework and avoiding overfitting. The study underscores the importance of integrating machine-learning models into practical cybersecurity applications. Future research should focus on improving these models and expanding their application across different domains to enhance cybersecurity defenses.

The exponential growth in technologies such as cloud computing, smart devices, virtualization, and the Internet of Things (IoT) has generated over four hundred zettabytes of network traffic data annually. This surge necessitates robust cybersecurity strategies to protect information from intrusions, which can result in significant financial losses. Reducing security risks requires the use of data analytics and machine learning to derive insights and make informed decisions based on network data. This study introduces the Flower Pollination Optimization (FPO) algorithm for feature selection to enhance the performance of several classifiers on the UNSW-NB15 dataset. We evaluated four classifiers: Linear Discriminant Analysis (LDA), Multi-Layer Perceptron (MLP), Quadratic Discriminant Analysis (QDA), and K-Nearest Neighbors (KNN) in two scenarios: without feature selection and with FPO-based feature selection. The results demonstrate significant improvements in classifier performance with FPO, with QDA achieving the highest accuracy of 99.16%. Comparative analysiswith recent studies highlights the superior performance of our approach, setting a new benchmark in intrusion detection. This research underscores the essential role of effective feature selection in improving the accuracy and reliability of Intrusion Detection Systems (IDS), particularly in IoT environments