An Intelligent Hybrid Technique of Decision Tree and Genetic Algorithm for E-Mail Spam Detection

Abstract

One of the most serious and irritating problem for decades is E-mail spam. These unwanted and unrequested commercial emails are also known as junk emails. Several machine learning approaches are used for detecting emails as spam or ham. These techniques detect spam emails from inbox and pass it to junk email folder. Even though among these approaches, it has been found that simple text classification techniques are not sufficient to detect spam e-mails. It is more preferable to use hybrid techniques to make detecting spam e-mails more efficiently. In this paper, a novel of hybrid machine learning technique of decision tree and genetic algorithm known as GADT is proposed for e-mail spam detection. It is believable that using genetic algorithm to improve the decision tree performance for text classification is effective and precise. Genetic algorithm is used to optimize and find the optimal value of a parameter named confidence factor that control the pruning of the decision tree. A major problem of any application of text classification, as spam detection, is its huge number of features that decrease in the accuracy of the classifiers. In our application, most of the extracted features that are extracted from the content of the email messages are irrelevant noise that can misled the classifier. So, it is found that dimension reduction stage is essential to reduce this huge number of features. In this paper, principle component analysis (PCA) technique is found to be a good choice to eliminate unsuitable features with less processing. The experimental results show the enhancement of accuracy of the hybrid approach GADT for detecting spam e-mail compared to the traditional decision tree. Also, these results show the high performance of GADT after using PCA compared to other traditional text classifiers.