A Novel Artificial Intelligence-Based Approach for Spam Email Detection Using Cross-Attention-Based BERT and Adaptive Convolutional Echo State Network

One of the most often used resources for communication is email, and millions of individuals and companies use it every day to exchange data. However, the surge in email users has been correlated with an unexpected rise in spam emails in modern days. Successful email extraction is becoming more challenging for both people and businesses. Furthermore, these communication channels are dangerous and have the potential to generate dangerous attacks known as spam. Email spam is one of the important issues for businesses and organizations, besides being a rising problem for normal internet users. Spam emails are unauthorized and undesired communications. These emails can be automatically identified using a variety of models and methodologies, but none of them has reached the highest level of accuracy. The main issues with email spam detection techniques are their high probability of false alarms and low detection rates. Therefore, this work utilizes the adaptive deep learning model for effective email spam classification. At first, the required text data from the email are collected through available data resources. Further, the feature extraction is performed through the Cross-Attention-based Multiscale Bidirectional Encoder Representations from Transformers (CA-MBERT), which can help to capture the relevant patterns. Subsequently, the Adaptive Convolutional Echo State Network (AC-ECN) is employed to effectively distinguish spam and non-spam emails. Here, the parameters in the proposed AC-ECN are optimized using the Intellectual and Randomized Hippopotamus Optimization Algorithm (IR-HOA) to achieve accurate results. Extensive comparative analysis is conducted, and the obtained outcome confirms the superior performance of the designed framework.