Rank based RELIEFG Method for Spam Mail Detection and Classification

Email plays a crucial role in day-to-day communication. An increase in spam emails poses a significant threat, leading to financial and resource losses. It's not easy to differentiate spam emails from legitimate emails due to large no.of features in datasets. Developing effective spam detection method is crucial for email security with missing values, outliers and noise. In the literature many methods have been existed among those rank-based Relief method is suitable for detection and classification. However, the relief method considers one hit and one miss. Due to this reason, it fails for multiclass classification. Due to the importance of Relief method, researchers have made many extensions to Relief. Among those ReliefF is the one of the latest variants of Relief method which improves accuracy by reducing the computational cost in the presence of missing values and noise for multiclass classifications. However, it fails in presentce of outliers. In this paper, a new variant ReliefG is proposed. This method reduces the computational cost by finding the highest rank of attributes and generating crisp set values by integrating efficient fuzzy membership function into a sigmoid function to improve the accuracy in the presence of outliers.