Dmitri Alperovitch

Identifying Image Spam based on Header and File Properties using C4.5 Decision Trees and Support Vector Machine Learning

Image spam poses a great threat to email communications due to high volumes, bigger bandwidth requirements, and higher processing requirements for filtering. We present a feature extraction and classification framework that operates on features that can be extracted from image files in a very fast fashion. The features considered are thoroughly analyzed regarding their information gain. We present classification performance results for C4.5 decision tree and support vector machine classifiers. Lastly, we compare the performance that can be achieved using these fast features to a more complex image classifier operating on morphological features extracted from fully decoded images. The proposed classifier is able to detect a large amount of malicious images while being computationally inexpensive.

Support Vector Machines and Random Forests Modeling for Spam Senders Behavior Analysis

Unwanted and malicious messages dominate email traffic and pose a great threat to the utility of email communications. Reputation systems have been getting momentum as the solution. Such systems extract email senders behavior data based on global sending distribution, analyze them and assign a value of trust to each IP address sending email messages. We build two models for the classification purpose. One is based on support vector machines (SVM) and the other is random forests(RF). Experimental results show that either classifier is effective. RF is slightly more accurate, but more expensive in terms of both time and space. SVM produces similar accuracy in a much faster manner if given modeling parameters. These classifiers can contribute to a reputation system as one source of analysis and increase its accuracy.