Abstract
<jats:p>A spike in email filtering due to a large number of detected emails leads to yearly losses. One way to mitigate this loss is to categorize different types of suspicious emails, such as fraudulent or promotional messages from unknown senders. The first steps in identifying message categorization were based on simple approaches, such as word filters. More complex methods, such as language modeling based on deep learning, are already being used. The text classification problem is often addressed using Recurrent Neural Networks (RNNs), with Gated Recurrent Units (GRUs) being a popular variant due to their efficiency in capturing sequential dependencies.. Since classifying phishing emails is the focus of this study, GRU techniques were used. This study's results show that, in a dropout-free environment, GRU attained a high accuracy rate. A large amount of detection mail is created worldwide from several botnets, which impacts the limited mailbox capacity. They affect the security of private mail and the loss of communication space. The time required to identify and reply to detected emails is affected by them. Identifying suspicious emails is still considered a challenging job in the modern day. Due to email detection frequency, identification can be improved. The researcher constructs a Gated Recurrent Unit- Recurrent Neural Network (GRU-RNN) to identify emails. A new method was tested with a Detecting basis dataset. The procedure is 99.8% accurate. After considerable testing, the researcher concludes that the offered technique detects emails well.</jats:p>