ARCHIVES

Email remains a vital communication medium, supporting seamless information exchange across personal and professional domains. However, its widespread use has also made it a prime target for cyber threats, with increasingly sophisticated and large-scale spam tactics compromising the security of government institutions, corporate infrastructures, and individual users. Traditional filtering systems are often inadequate in addressing these evolving challenges. This study introduces an intelligent system designed to classify email messages into four specific categories: normal, fraudulent, harassment, and suspicious. The proposed solution combines advanced language processing techniques with a bidirectional neural network model capable of understanding contextual and semantic patterns within email content. A sample augmentation strategy is employed to enrich the diversity of training data, enhancing the model’s ability to generalize across varied input types. A rigorous evaluation process confirms the system’s robustness and accuracy in handling complex and high-volume datasets. Experimental results reveal a significant improvement in classification performance, with the system achieving an accuracy rate of 99.1 percent. The architecture effectively supports digital forensic applications by extracting and analyzing meaningful content from diverse communications. This research contributes a reliable and scalable method for email threat detection, bolstering cyber security efforts in an increasingly interconnected digital landscape.