Feature extraction for multi-label learning in the domain of email classification

Carmona-Cejudo, José M.; Baena-García, Manuel; Campo-Ávila, José del; Morales-Bueno, Rafael

doi:10.1109/cidm.2011.5949301

Cited by 8 publications

(5 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The analysis of cyber-attacks is a must for detecting the attacker and preventing their further malicious activities. The digital information security forensics is a time-and resource-consuming process, therefore automation should be used as much as possible to reduce the investigation time as well as to increase its accuracy [4,5]. One of the first steps in the forensics is classification of obtained data and its prioritization.…”

Section: Introductionmentioning

confidence: 99%

Multi-Language Spam/Phishing Classification by Email Body Text: Toward Automated Security Incident Investigation

et al. 2021

View full text Add to dashboard Cite

Spamming and phishing are two types of emailing that are annoying and unwanted, differing by the potential threat and impact to the user. Automated classification of these categories can increase the users’ awareness as well as to be used for incident investigation prioritization or automated fact gathering. However, currently there are no scientific papers focusing on email classification concerning these two categories of spam and phishing emails. Therefore this paper presents a solution, based on email message body text automated classification into spam and phishing emails. We apply the proposed solution for email classification, written in three languages: English, Russian, and Lithuanian. As most public email datasets almost exclusively collect English emails, we investigate the suitability of automated dataset translation to adapt it to email classification, written in other languages. Experiments on public dataset usage limitations for a specific organization are executed in this paper to evaluate the need of dataset updates for more accurate classification results.

show abstract

Section: Introductionmentioning

confidence: 99%

Multi-Language Spam/Phishing Classification by Email Body Text: Toward Automated Security Incident Investigation

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Notwithstanding the flexibility and applicability of neural networks to this problem, they rely on extracted features as the inputs to their decision engine, like all machine learning algorithms. Much research has been done on feature extraction, with some important conclusions contained in [13,14,15,16,4,17,18,19,7,20]. The research discussed in this paper uses a similar approach to that discussed in [10], using a combined set of both statistical and semantic features.…”

Section: Preliminariesmentioning

confidence: 99%

Semantics based multi-layered networks for spam email detection

Creech

Jiang

2012

AIP Conference Proceedings

View full text Add to dashboard Cite

Articles you may be interested inQuantum control of a model qubit based on a multi-layered quantum dot Abstract. In this paper, we propose a hybrid semantic and statistics approach for spam email detection. An adaptive training scheme is implemented, which does not require a large training pool. Experimental results have shown promising performance. INTRODUCTIONUnsolicited e-mail has become the bane of modern life. The impact of spam emails on business, government, and general activity ranges from irritating to severely deleterious. Spam has two main effects: first, it clogs the Internet, reducing effective throughput and slowing the transmission of genuine communications; second, there is significant aggregate time wasted in deleting such e-mails. Whilst it may only take a few seconds for one individual to remove one spam e-mail, the total time squandered when this same e-mail is deleted by millions of recipients becomes significant, with consequent reduction in productivity and increasingly drain on IT resources. In an attempt to combat the effect of spam, numerous email filters are currently deployed on most operating systems and ISPs. Tuning these spam filters is a complex task, as the impact of a false positive, and the consequent blocking of the bona fide e-mail, has a much greater impact on the individual user than a false negative and the subsequent transmission of a spam e-mail. End users are hence reluctant to use overly aggressive spam filters, regardless of the overall impact on Internet traffic. Significant gains in global efficiency could be provided by a highly accurate spam filter.This paper introduces a hybrid approach to the spam problem, utilising a combined semantic and statistical text based feature extraction methodology, it is coupled with three multilayer feedforward back propagation neural networks operating in a combined voting architecture. A level of self training was implemented using high confidence unanimous decisions to further develop the network beyond its initial training set. Significantly, good accuracy was achieved on previously unseen e-mail test cases using only 11 features and an initial training set of only 200 elements. The adaptive nature of this system coupled with the small initial training set suggests that a full implementation would perform well in dynamic environments, with the neural network topology lending itself to possible hardware implementations as explored in [1,2]. The voting methodology adopted by this algorithm significantly mitigates the training overhead traditionally required by neural networks, as each network in this approach is only required to find a local minima, rather than converging to the elusive global result of a perfect system. Errors arising in any one neural network are generally accounted for by correct classifications from the other two networks, resulting in an aggregate system with a high accuracy rate, low initial training requirements, and a self reinforcing adaptive learning capability. This paper is organised as follows: in Curr...

show abstract

“…Therefore, the model for predicting mass flow by using the feed rate of fuel is not static and it is a cause of concept drift. Another example of concept drift is a spam mail filtering [3,4]. The main source of concept drift in spam mail filtering are the changes in e-mail content and presentation, an adaptive behavior of spammers (may be virtual drift or real drift), and user's attitude that may change the categories of spam and legitimate e-mails (real drift).…”

Section: Introductionmentioning

confidence: 99%

Concept drift for CRD prediction in broiler farms

Channoi

Maneewongvatana

2015

2015 12th International Joint Conference on Computer Science and Software Engineering (JCSSE)

View full text Add to dashboard Cite

Chronic Respiratory Disease (CRD) is a serious problem in broiler farms and food production industry. This disease cannot be observed easily during the broiler raised process. The model for predicting CRD rate is not exactly identified, because of the variation in farm environment and the development of breeding. Therefore the embedded of concept drift to the normal predictor is the possible way to build the system that can detect the change in data stream and build the suitable predictor for the current information. In this work we applied the method of concept drift to the basic Naïve Bayes classifier to predict the CRD rate by using the data collected from many contract farms during the entire raising period. The experimental result shows that, with the concept drift embedded method, the accuracy of prediction is improved by about 30%.

show abstract

Feature extraction for multi-label learning in the domain of email classification

Cited by 8 publications

References 15 publications

Multi-Language Spam/Phishing Classification by Email Body Text: Toward Automated Security Incident Investigation

Multi-Language Spam/Phishing Classification by Email Body Text: Toward Automated Security Incident Investigation

Semantics based multi-layered networks for spam email detection

Concept drift for CRD prediction in broiler farms

Contact Info

Product

Resources

About