AN AI-DRIVEN HYBRID MODEL FOR CYBERSECURITY THREAT DETECTION USING AUTOENCODER AND LSTM

Abstract

Cybersecurity threats are evolving rapidly, requiring advanced detection mechanisms beyond traditional rule-based systems. This study proposes an AI-driven hybrid model integrating Autoencoder and Long Short-Term Memory (LSTM) networks to enhance real-time threat detection. The Autoencoder extracts hidden patterns and reduces dimensionality, while LSTM captures sequential attack behaviors for better classification. The proposed model effectively reduces false positives and negatives, addresses data imbalance issues, and classifies threats based on severity (Low, Medium, High). Performance evaluation using accuracy, precision, recall and F1-score demonstrates the model’s reliability in detecting sophisticated cyber threats. The results confirm the potential of deep learning in improving cybersecurity resilience and real-time threat mitigation.