Western Technical College

Denial-of-Service (DoS) attacks are still a serious threat to cloud computing systems and the large volume of malicious traffic can deteriorate network performance and cause service unavailability. Conventional intrusion detection systems (IDS) are mostly based on packet-content features and have difficulty to identify temporal traffic patterns, especially in the imbalanced traffic scenarios. This paper presents a time-sensitive ID framework that uses packet IAT and traffic features to achieve efficient detection of DoS in the cloud. Specifically, a Mininet integrated dataset was created comprising normal data, ICMP flood traffic (attack type 1), TCP flood traffic (attack type 2), and Smurf traffic (Other) from which we built nine models using traditional machine learning and deep learning approaches. To avoid the influence of imbalanced class and improve learning results, we utilized Synthetic Minority Oversampling Technique (SMOTE) technology and Grid Search (GS) optimization. Experiment results show that although traditional methods are able to obtain moderate performance, deep learning models achieved superior macro-averaged performance across all classes and reach 97 % overall accuracy as well as nearly perfect precision and recall in all types of attacks (including the challenge case, the Smurf traffic) after applying our RNN-SMOTE-GS model. The findings demonstrate that the use of spatio-temporal traffic behavior information, as well as optimization strategies, demonstrates the effectiveness of temporal modeling combined with imbalanceaware optimization our detection capability. This platform offers the necessary underpinnings for real-time intrusion detection in cloud computing environments.

Ransomware is a major cybersecurity problem with profound impacts on data damages and business operations. Fast generation of new ransomware families has undermined the accuracy of remote traditional signature-based and singlemodel mechanisms for detection, mainly into multi-class malware categorization. To meet these challenges, an ensemble machine learning-based multi-layer ransomware detection procedure using feature selection on static behavior data is presented in this paper. The proposed framework employs a feature selection technique using Random Forest to reduce dimensionality and capture discriminative features. We employ two base learners: (1) an MLP for learning global non-linear feature interactions, and (2) a 1D CNN that learns local patterns. Their predicted probabilities are stacked with a Random Forest meta-learner for enhanced robustness and generality. Performance of the model is measured on 5 classes (Backdoor, Dropper, Ransomware, Trojan and Benign) from CIC-AndMal-2020 dataset. Experiments demonstrate that the proposed method has an accuracy of 99.81 % in multiclass classification which is better than existing methods. Crossvalidation and ROC-AUC analysis further verified the stability, reliability, and potential of our model in real-world ransomware detection.

Phishing remains one of the most pressing security problem with increasingly sophisticated attacks that are difficult to detect by existing systems. In this paper, we propose a new hybrid phishing detection system named as Balanced Deep Neural Network (BDNN) to deal with class imbalance issue and to enhance the detection sensitivity achieved through a mixture of Machine Learning (ML) and Deep Learning (DL) models. We propose BDNN that uses weighted loss functions and dynamic sampling to promote detection performance, especially for minority locating phishing examples. Leveraging the merits of ML models for structural URL feature extraction and DL models for learning sequential patterns, our approach could reach high accuracy of 96.5% and recall of 96.6%. The system is deployed as a real-time browser extension and less than 200 ms to classify URLs, guaranteeing users smooth experience. Additionally, it contains an ongoing learning component that is used to counter new phishing strategies. Experimental results demonstrate that the proposed BDNN ensemble not only outperforms some state of the art traditional models, but also has a high ease of deployment for real-time applications.

The practice of English language teaching has long been an important part of socializing transnational migrants and international students into ongoing English-dominant settler-colonial projects in North America and beyond. The professional activities, knowledge, and identities of English language instructors are therefore central to the reproduction of the settler-colonial order. In this article, I investigate the relationship between language-teacher identities and settler-colonial discourses of raciolinguistic differentiation and hierarchy in Canada. Working in a discursive constructionist conceptual framework, I adopt occasioned semantics to analyze excerpts from research interviews with two ELT instructors in post-secondary and adult ESL contexts. I demonstrate how these instructors' talk about students and languages performed language-teacher identities-in-discourse and argue that these performances reflected and contributed to reproduction of settler-colonial discourse patterns. I further suggest that settler colonialism constitutes for ELT practice in Canada a hermetic "universe" with its own internal logics and relations that must be examined and made explicit through reflection. Some pedagogical implications of this analysis include the need for ELT instructors as well as English-language teacher educators to develop an awareness of local settler-colonial histories, teach for truth and reconciliation, and embrace strategies for de-naturalizing the settler-colonial "universe" as they create spaces of possibility for decolonization to be carried out. One research implication is that language-teacher identity scholarship would be strengthened by embracing epistemological and methodological decolonization.

This article is a review of Jiř ubrts book The Sociology of Time: A Critical Overview (Cham: Palgrave Macmillan/Springer; 2021. 283 p.). The author places this work in a broader context of previous books by ubrt in order to show that all these publications analyze the past and contemporary sociological theories, and focus on historical sociology and the conception of sociology as a science on social processes. ubrt considers time in the context of the long-term development of knowledge, in which efforts have been made to control and master it. He also conducts a critical analysis of the views of previous generations of sociologists who developed ideas about the nature and functions of time. ubrt examines different fields of the so-called sociology of time; however, his main interest is the temporalized sociology such as theories of Niklas Luhmann and Anthony Giddens, but especially the conceptions in the field of historical comparative sociology, which combine the object of sociological research with long-term historical processes. According to ubrt, the basic aspect of time that should be decisive for sociology is its irreversibility associated with the idea of an open future.