Yazar "Abduljabbar, Elaf Talib Abduljabbar" seçeneğine göre listele

Listeleniyor 1 - 1 / 1
  • Küçük Resim
    Öğe
    ANALYSIS OF MACHINE LEARNING AND DEEP LEARNING TECHNIQUES FOR RANSOMWARE DETECTION
    (2023-10) Abduljabbar, Elaf Talib Abduljabbar
    Ransomware is a special type of malware by which the attacker targets the victim's device using a link attached by an email, and once the victim opens the attachment, all his files are encrypted. The victim cannot retrieve his data without paying the attacker for the decryption key. Ransomware becomes very dangerous and affects all human facilities, including medical centers, military organizations, security platforms, financial facilities, etc. Ransomware detection and classification-based artificial intelligence applications are essential to limit the attacker's ability and prevent it from harming devices. The current study proposes a new ransomware detection and classification study. Besides, a novel feature selection algorithm is proposed to involve the essential information of network tasks and drop the redundant data that can slow the detection process. The study uses a challenging dataset of 392034 records, 84 features, and 11 different types of ransomware. In the first step, the dataset is preprocessed by cleaning it, encoding all textual (categorical) features, and normalizing them to ensure it fits all machine learning and deep learning models. In the second step, the dataset is split into the train (80%) and test (20%) for the machine learning models. Besides, another validation set is created with a percentage of (20%) of the training set for the deep learning models. The third step is feature selection, in which the redundant features are dropped using a novel hybrid feature selection method depending on both ANOVA and Random Forests to select the best subset of features. In the fourth step, many machine learning and deep learning models are trained using the training set. The experiment part includes applying the fusion of the individual models (for both machine learning and deep learning models) besides the ensemble learning of these individual models. In the evaluation step, the precision, recall, F1-score, and accuracy are used to assess the performance of the individual, the fusion, and the ensemble models. Besides this, three different feature selection scenarios are conducted to seek the best combination of features. The training time of all models is also computed to see the effect of the feature reduction on the computational costs. Results showed that the best models are the XGB, LGBM, and RF models. Besides that, the ML ensemble model achieves a good performance. The feature selection method minimized the training time significantly, especially for the high-computational models like XGB and LGBM, without any remarkable degradation in performance. The best-obtained accuracy is related to the XGB model with 99.87%. The study is also compared with the current state-of-art methodologies. The comparison proves that the current study outperforms all previous ones. Future work can focus on the idea of hyperparameter optimization to improve the performance.