Adversarial Spam Detection Using the Randomized Hough Transform-Support Vector Machine

DeBarr, Dave; Sun, Hao; Wechsler, Harry

doi:10.1109/icmla.2013.61

Cited by 17 publications

(12 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(1) KPCA 4 and CSVM 5 are employed by Geng et al (2014), where feature selection is the duty of KPCA and CSVM works as the classi¯cation algorithm, while it achieved a higher detection rate and improved detection e±ciency. (2) RHS 6 and SVM 7 are proposed by Debarr et al (2013) that showed two combined algorithms are su±cient to improve 9.3% versus RONI. (3) Combination of NSA 8 and various evaluations in another article by Idris and Selamat (2014) are proposed.…”

Section: Related Workmentioning

confidence: 99%

Classification Spam Email with Elimination of Unsuitable Features with Hybrid of GA-Naive Bayes

EBADATI

Ahmadzadeh

2019

J. Info. Know. Mgmt.

View full text Add to dashboard Cite

Email spam is a security problem that involves different techniques in machine learning to solve this problem. The rise of this security issue makes organisation email service unreliable and has a direct relation with vulnerability of clients through unexpected spam mails, like ransomware. There are several methods to identifying spam emails. Most of these methods focused on feature selection; however, these models decreased the accuracy of the detection. This paper proposed a novel spam detection method that is not only to decrease the accuracy, but eliminates unsuitable features with less processing. The features are in the terms of contents, and the number of features is very big, so it can decrease the memory complexity. We use Hewlett-Packet (HP) laboratory samples text emails. First, GA algorithm is employed to select features without limited number of feature selection with the aid of Bayesian theory as a fitness function and checked with a different number of repetitions. The result improved with GA by increasing number of repetitions, and tested with distinctive selection method, Random selection and Tournament selection. In the second stage, the dataset classifies emails as Spam or Ham by Naive Bayes. The results show that Naive Bayes and hybrid GA-Naive Bayes are almost identical, but GA-Naive Bayes has a better performance.

show abstract

Section: Related Workmentioning

confidence: 99%

Classification Spam Email with Elimination of Unsuitable Features with Hybrid of GA-Naive Bayes

EBADATI

Ahmadzadeh

2019

J. Info. Know. Mgmt.

View full text Add to dashboard Cite

show abstract

“…This leverages consensus reasoning (e.g., RANSAC) and helps with both adversarial learning and clonal selection. The motivation behind re-identification comes from the fact that the multitude of antigens share common characteristics and can be traced to some common source(s) using amongst others longest common sequence (LCS) (e.g., positive selection) for similarity using dynamic programming and/or RANSAC/Random Hough Transform for realization [7]. Positive selection further helps with better designs for adversarial attacks as they anticipate weak points in defense susceptible to be overwhelmed by persistent attacks.…”

Section: Immunity and Detectionmentioning

confidence: 99%

“…This suggests that the defense should exercise great caution in its use of training data. Towards that end, labeling errors characteristic of lacking proper annotation can be detected and redressed in many ways including constrained regularization driven optimization and label flipping [6] with label flipping best done in a principled way characteristic of importance sampling [7]. The use of soft rather than hard labeled data is more resistant to (adversarial) label noise [3] with soft labels integral to conformal prediction, the learning framework proposed here, and to incremental transduction, in particular [8].…”

Section: Introductionmentioning

confidence: 99%

“…It then iteratively enlarges the set with consistent data when possible and re-estimates the model and its error. Starting from different subsets, The Randomized Hough Transform-Support Vector Machine (RHT-SVM) [7] leverages multiple versions of the decision boundary to identify messages that have been mislabeled deliberately, as result of persistent attacks, or not deliberately but due to poor Oracle annotation. The RHT-SVM uses the product of the actual classification label and the average signed distance of an observation from the decision boundary to determine if a training message has been mislabeled.…”

Section: Semantics and Stratificationmentioning

confidence: 99%

“…The motivation here is to confront attack diversity and scale using semantics and stratification using clustering (e.g., Chinese Restaurant Process and spectral clustering), topic discovery using probabilistic latent semantic analysis (PLSA) and latent Dirichl et al location (LDA)) [37], and aggregation and consensus reasoning using RANSAC and Random-Hough Transform) [7]. Defense can be further enhanced using feature reputation [17].…”

Section: Semantics and Stratificationmentioning

confidence: 99%

See 2 more Smart Citations

Cyberspace Security Using Adversarial Learning and Conformal Prediction

Wechsler¹

2015

IIM

Self Cite

View full text Add to dashboard Cite

This paper advances new directions for cyber security using adversarial learning and conformal prediction in order to enhance network and computing services defenses against adaptive, malicious, persistent, and tactical offensive threats. Conformal prediction is the principled and unified adaptive and learning framework used to design, develop, and deploy a multi-faceted self-managing defensive shield to detect, disrupt, and deny intrusive attacks, hostile and malicious behavior, and subterfuge. Conformal prediction leverages apparent relationships between immunity and intrusion detection using non-conformity measures characteristic of affinity, a typicality, and surprise, to recognize patterns and messages as friend or foe and to respond to them accordingly. The solutions proffered throughout are built around active learning, meta-reasoning, randomness, distributed semantics and stratification, and most important and above all around adaptive Oracles. The motivation for using conformal prediction and its immediate off-spring, those of semi-supervised learning and transduction, comes from them first and foremost supporting discriminative and non-parametric methods characteristic of principled demarcation using cohorts and sensitivity analysis to hedge on the prediction outcomes including negative selection, on one side, and providing credibility and confidence indices that assist meta-reasoning and information fusion.

show abstract

SecMD: Make Machine Learning More Secure Against Adversarial Malware Attacks

Chen

2017

AI 2017: Advances in Artificial Intelligence

View full text Add to dashboard Cite

Adversarial Spam Detection Using the Randomized Hough Transform-Support Vector Machine

Cited by 17 publications

References 13 publications

Classification Spam Email with Elimination of Unsuitable Features with Hybrid of GA-Naive Bayes

Classification Spam Email with Elimination of Unsuitable Features with Hybrid of GA-Naive Bayes

Cyberspace Security Using Adversarial Learning and Conformal Prediction

SecMD: Make Machine Learning More Secure Against Adversarial Malware Attacks

Contact Info

Product

Resources

About