Email system serves as one of the most powerful com- munication systems for transmitting the user’s information from one to another. It includes not only text but also images, files, etc. This platform helps the user in saving a huge amount of time and money in comparison with out- dated techniques like telegrams, etc. Nowadays, approxi- mately 281 billion emails are transmitted all over the world in our day-to-day life (https://cacm.acm.org/magazines/ Table 1 Comparative study Fig. 1 Layout of proposed method Table 2 Feature selection schemes the total number of mails, w is the word, P(H) is the probability for ham feature and P(S) is the probability for spam feature. This is clearly shown in Eq. (5) and Eq. (6), respectively. Fig. 4 Comparative study of different machine learning algorithms. E Enron dataset, LS Ling-Spam dataset E Enron dataset, LS Ling-Spam dataset, Info Gain information gain, SBS sequential backward selection, EFS exhaustive feature selection, SFS sequential forward selection, BFS best first search, WFS wrapper- based feature selection RF is the best among the rest classifiers in the presence of with and without the strategies used for selection of features. From the outcomes, it tends to be seen that with the strategies used for selection of features such as sequential feature selector, GridSearchCV and exhaustive feature selector, best first search has performed in a good manner in comparison with others—aside from the NB in the Enron dataset E Enron dataset, LS Ling-Spam dataset To judge the performance of the algorithm, the two important questions arises here are: (1) Are the features defined in a well manner? (2) Do the amount of data required during the training phase are sufficient? To answer such questions, the possible solution is to check the size of the dataset during training time. With the increase in the size of training data, it becomes very much complex for a Fig. 5 Four cross-fold validation techniques. A—trained features, B—test features, S()—selection() and TS—training set
