Malicious E-Mail Detection using Artificial Neural Networks

Abstract

The malicious email creates troubles for computer users and evolving threatsmainly for Internet users. Some of the possessions of spam are itcorrupts Inbox with manyridiculous emails.It is a root cause ofdegrading the Internet speed to someamount. To combat this, different data mining classifiers developed to detectmaliciousmail. The present paper attempts to classify spam emails viathe machine learning approach. The very purpose of this research is to classifyham and spam mails by constructing a sensitive classification modelwith a low false-positive rate. Three noteworthy machine learning algorithms namely, Artificial Neural Network (ANN), Decision Tree and J48 are studied for their efficiency for email classification. In the ANN model, multilayer perception is trained with backpropagation learning algorithm for weight adjustments for hidden layers. We employed supervised machine learning techniques to filter the email spam messages. We import data from different g-mail accounts and applied preprocessing techniques. In the end, the three algorithms are compared according to their evaluation criteria. Because false positives are more expensive than false negatives, we obtained the predictive accuracy of the classifier with the help of false negatives and false positives.The results of the multilayer perceptron are sound enough when compared to its accuracy,false positive & false negative rates, precision, recall, receiver operating characteristic curve area and precision-recallcurve area. Finally, we present the comparison of three classifiers with visualization techniques and obtain the predictive accuracy of the classifiers. All thesimulated values arepresented at the endof thepaper using various visualization techniques.

Keywords:Artificial Neural Network (ANN), Precision, Recall, False positives (FP), False negatives (FN), and Receiver Operating Characteristics (ROC)