|
|
|
 |
Search published articles |
 |
|
Showing 2 results for Moghaddam
Dr Majid Fani, Dr Mohammadamin Torabi, Dr Matineh Moghaddam,
Volume 9, Issue 2 (2-2021)
Abstract
Not all phishing attacks are always done in the form of website forgery and telephone phishing. Emails and messages that appear to be sent by the bank and receive information from the user can also be a phishing attack. Feature selection and sample selection are two very important issues in the data processing stage in detecting malicious emails. In particular, identifying spam without data reduction will not be nearly as accurate in the results. Most articles and research have focused on one of these issues, and there are few articles that have worked in combination to detect malicious emails. Therefore, the purpose of the present study is to provide a method to reduce the data in identifying emails by selecting features and samples simultaneously. In the proposed method in this paper, the forbidden search algorithm and the genetic algorithm are used in combination and simultaneously. For the suitability of this method, the evaluation vector machine evaluation function was used. The results showed that the detection rate of spam and e-mails in LineSpam and UCI datasets was 97.28, which was the highest possible value compared to other algorithms proposed in previous studies.
Mahmoud Saeidi, Nasrin Taaj, Azadeh Bamdad Moghaddam,
Volume 11, Issue 2 (3-2023)
Abstract
The proposed method for implementing cryptographic traffic detector systems and evasive tools in this article is a method based on deep learning, which due to the importance of optimal and automatic extraction of features from the input data set, an automatic encoder network has been used in the feature extraction phase. Then, the output of the middle hidden layer of this network is applied to a deep convolutional neural network. Deep convolutional neural networks can play an effective role in improving the performance of the system by taking into account the spatial connections between features. Finally, the output of deep convolutional neural network is also applied to two fully connected layers in order to perform the classification process. So that the number of neurons in the second fully connected layer will be equal to the expected number of layers of the system. In the first part of this article, the specifications of the proposed systems for the implementation of cryptographic traffic detectors and evasion tools will be mentioned in terms of operational architecture and features. Then, the characteristics of the networks used in the implementation of the system, and the method of their integration to form the final identifier system are briefly stated. After that, the training stage of the system and its implementation method will be introduced and at the end, how to adjust the parameters of the model and the mechanisms used to improve the overall performance of the system and the results of its performance evaluation will be presented.
|
|
