A Probabilistic Neural Network Based Classification of Spam Mails Using Particle Swarm Optimization Feature Selection

Email has gained the explosive growth in the communication of people across the world. This worldwide communication also has some disadvantages like Spam mails. The spammers spread the useless, unwanted mails and even malicious contents to the usersemails. This increasing number of spam mails increases the need for the spam detection architecture with the machine learning classification. The proposed spam detection architecture composed of a feature selection process to minimize the error rate, a redundancy removing method and finally a classification system for categorizing the spam mails from the legitimate mails. The incoming mails are preprocessed by using the three traditional steps such as Tokenization, Stemming and the Stop Word Removal. The Vector Quantization (VQ) process is utilized to remove the redundancy in both the training and preprocessed data. Then the preprocessed redundancy removed training and testing data are given to the feature selector called the familiar Particle Swarm Optimization (PSO) algorithm which mines the optimal features suitable for the classification. Finally, along with the selected features, the Probabilistic Neural Network (PNN) classifies the spam mails from the legitimate mails with more accuracy and precision.