Classification of Threats Using Deep Learning Algorithms

Cybersecurity threats are becoming more complex and frequent, making traditional detection methods insufficient. To address this, the project introduces a deep learning-based system for automatically detecting and classifying network threats. It uses structured traffic data from the CyberFedDefender dataset to train two neural network models. The first model performs binary classification to determine whether a traffic record is benign or malicious. If a threat is detected, the second model performs multiclass classification to identify the specific type of attack, such as DoS, DDoS, or PortScan.Both models are fully connected feedforward neural networks that use ReLU activation for efficient training and dropout for regularization. The data is preprocessed with label encoding for categorical fields like Protocol and Flags, and standardized using a scaler for numerical consistency. The complete system is deployed through a Streamlit web interface. Users can input traffic data via form or upload CSV files for batch analysis. Results are displayed in an interactive table with options for downloading and visualization. This framework shows how deep learning can enhance the speed, accuracy, and scalability of cyber threat detection and classification