Detecting attacks on e-mail

E-mail has become a popular communication tool widely used by universities, enterprises and governments. Despite the convenience it brought to people, attacks on e-mail happen very frequently in the range of the world, causing large economic loss and occupying a mass of network bandwidth every year. The hazards from e-mail attacks underline the importance of detecting and resisting spam in an efficient and timely way. Using Python, we built Naive Bayes (NB) and support vector machine (SVM) filters for emails. The filtering performance of NB and SVM email filters applying different kernel functions was compared and evaluated based on several evaluation indices including accuracy, precision, and total cost ratio (TCR). Also, in order to optimize the filters, the influences of stop words removal, feature numbers and other parameters in the filtering algorithms were monitored.