Distributed Denial of Service (DDoS) attacks present a significant challenge to the reliability and security of IoT networks, often resulting in service disruptions, data breaches, and operational downtime. This research introduces an advanced deep ensemble learning framework for detecting DDoS attacks, enhanced with Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. CNNs are employed to extract spatial correlations from network flow features, while LSTMs capture sequential temporal dependencies in traffic behavior, enabling accurate modeling of complex attack patterns.

The proposed framework integrates deep learning with traditional ensemble classifiers, including a Stacking Classifier (Random Forest, Gradient Boosting, Naive Bayes), a Voting Classifier (Logistic Regression, Decision Tree, KNN, AdaBoost), and a TPOT Classifier that automates pipeline optimization. Additionally, pruning techniques are applied to remove redundant or low-performing models, improving computational efficiency.

A comprehensive set of network traffic features including packet lengths, inter-arrival times, flow duration, and TCP flags are used for training. The system is evaluated using accuracy, precision, recall, and F1-score, demonstrating high detection accuracy and low false positives. The combined use of CNN, LSTM, and optimized ensembles makes the approach scalable and robust for real-time DDoS detection in IoT environments.

Keywords: DDoS detection, IoT security, CNN, LSTM, ensemble learning, TPOT, stacking, pruning, traffic analysis, network features.