Multi-Feature Hybrid LSTM-CNN Framework for Phishing Email Detection

Phishing has become one of the most universal cyber-attacks, leveraging users' trust to hijack sensitive information such as login credentials, financial data, and personal information. With the increasing sophistication of phishing techniques, traditional rule-based methods and signature-based detection approaches have become inadequate. This study proposes an advanced phishing email detection system on multiple datasets using a hybrid deep learning method that incorporates Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN). The research methodology consists of the following steps: Dataset Collection and Pre-processing, Feature Extraction, and Hybrid LSTM-CNN Model Architecture. A feature extraction phase enhances detection by incorporating email metadata, URL patterns, and embedded links. The model hybrid LSTM-CNN model achived the highest accuracy in three different datset like Enron 98.2%, SpamAssassin 97.5% and Kaggle phishing email dataset 96.8% than LSTM, CNN, SVM, Random Forest and BERT. Apart from accuracy, this model also gained the highest score in precision 97.9%, recall 98.5% and F1-score 98.2% in critical evaluation metrics. This approach demonstrates the efficiency of deep learning methods for phishing detection attempts and enhancing email security. Furthermore, the proposed system can be incorporated into electronic communication devices such as secure email servers, smart gateways, and IoT-based communication devices. With its integration of detection technology into electronics hardware, firmware, and network protocol design, the system allows for real-time threat prevention, reduces network vulnerabilities, and maximizes the reliability of modern communication infrastructures