Deep Learning Models for Anomaly Detection in Autonomous Network Troubleshooting and Optimization: A Comparative Study
Abstract – To identify anomalies in network traffic patterns through the NSL-KDD datasets, this study created and tested six deep models: Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Transformer. The main aim was to evaluate the performance of these models to detect network anomalies as this is critical in providing the security and integrity of autonomous network systems. The data preprocessing phase comprised of normalization, splitting of data, and dimensionality reduction using Principal Component Analysis (PCA). All models were trained and tested using high-performance metrics like accuracy, precision, recall, and F1-score with libraries like TensorFlow, Pandas, Matplotlib, NumPy, and Scikit-learnand. According to the results, BiLSTM performed better. The findings showed that: BiLSTM performed better than other models with precision and recall rates of 98.43% and 98.13%, respectively. GRU was second with 95.51% accuracy, followed by LSTM (94.15%) and CNN (92.78%). Surprisingly, the Transformer model performed very poorly with 6.00% accuracy. These findings resonate with previous studies conducted by Imran et al., who used BiLSTM and achieved 91.36% accuracy, and Li et al., who achieved 82.87% accuracy using GRU on NSL-KDD. This work is a useful study on using deep learning models for network security and verifies the performance of the BiLSTM model for anomaly detection. The future work will explore further improvement of model performance and adjustment to new network threats.