Authors: Assistant Professor M.Swathi, Kolipaka Aishwarya
Abstract: The rapid growth of instant messaging platforms has significantly improved digital communication while simultaneously increasing the spread of unsolicited and malicious messages. Spam messages not only interrupt user interactions but also expose individuals to phishing attacks, malware, financial scams, and privacy risks. and struggle to capture contextual relationships within sequential text, limiting their effectiveness against evolving spam patterns. To overcome these limitations, this research presents a real-time spam message detection sequence padding, and word embedding to transform raw chat messages into structured numerical representations suitable for deep learning. The LSTM network is trained using a labelled SMS spam dataset and optimized with the Adam optimizer and binary cross-entropy loss function. To improve model generalization and reduce overfitting, dropout regularization and early stopping mechanisms are incorporated during training. The trained model is integrated into a chat application where every outgoing message is automatically analyzed before delivery. Experimental evaluation demonstrates that the proposed framework effectively distinguishes spam from legitimate messages while maintaining high prediction accuracy and reliable real-time performance. The system successfully combines intelligent spam filtering with standard messaging functionalities such as individual conversations, group chats, multimedia sharing, and user authentication, providing a secure and user-friendly communication environment.
