Nowadays, we use emails almost in every field; there is not a single day, hour, or minute when emails are not used by people worldwide. Emails can be categorized into two types: ham and spam. Hams are useful emails, while spam is junk or unwanted emails. Spam emails may carry some unwanted, harmful information or viruses with them, which might harm user privacy. Spam mails are used to harm people by wasting their time and energy and stealing valuable information. Due to increasing in spam emails rapidly, spam detection and filtering are the prominent problems that need to be solved. This paper discusses various machine learning models like Naïve Bayes, Support Vector Machine, Decision Tree, Extra Decision Tree, Linear regression., and surveys about these machine learning techniques for email spam detection in terms of their accuracy and precision. In this paper, a comprehensive comparison of these techniques and stacking of different algorithms is also made based on their speed, accuracy, and precision performance.

On the basis of characteristics derived from IPv4 addresses, this paper offers a method for identifying interaction linked with website-based malware and then modelling a machine-learning-based classifier.  In this research work, a modified approach is proposed for detecting fraudulent websites and compared with other methods like SVM assessment of IP addresses, octet-based technique, modified extended version of octet-based technique, and bit string-based characteristics. This modified approach is based on the fact that logical addressing is more reliable and consistent than other measures like URLs and DNS. The characteristic sequence which makes up URLs and domain names are more changeable with respect to IP addresses which are less changeable in comparison to URLs or domain names. The IPv4 address length is encoded into 4-byte space. Here, we have evaluated our modified approach with valid IP addresses from Kaggle [11], published on January 16, 2018, have been used to validate the efficacy of their metho.

Question classification plays vital role in Question Answering (QA) systems. The task of classifying a question to appropriate class is performed to predict the question type of the natural language question. In this paper, initially we have presented a brief overview of classification approaches adapted by different question answering systems so far and then propose a two-way question classification approach for higher education domain which not only identifies focus word and question class but also reduces answer search space within corpus comprise of question-answer pair, adding to the classification accuracy. For precise semantic interpretation of domain keywords, a domain specific dictionary is constructed which primarily have four domain word type. Classified features are built upon domain attributes in the form of constraints. The experiment proved the efficiency for restricted domain, even though we used quite simplistic approach.