A Lightweight 1D-CNN for Edge-Based Intrusion Detection in IoT Networks

The rapid proliferation of Internet of Things (IoT) devices has revolutionized connectivity but introduced significant vulnerabilities in data transmission, particularly during transit where eavesdropping, man-in-the-middle attacks, and denial-of-service threats compromise confidentiality and integrity. This study proposes a lightweight convolutional neural network (CNN) model for real-time anomaly detection in IoT network traffic to ensure secure data transit. The methodology utilizes the IoTID20 benchmark dataset, comprising over 625,000 instances of labelled traffic from simulated smart home environments with Wi-Fi cameras and routers. Data preprocessing involves principal component analysis for dimensionality reduction, followed by train-validation-test splits (70%-15%-15%). The CNN architecture employs one-dimensional convolutions to capture temporal patterns in packet sequences, with batch normalization, dropout for regularization, and Adam optimization. Training incorporates class weighting to address imbalance and data augmentation for robustness. Evaluation metrics demonstrate superior performance, achieving 95% accuracy, 97.7% precision, 97.1% recall, and 97.4% F1-score on test data, outperforming an autoencoder baseline (84% accuracy). Confusion matrix analysis reveals minimal false positives (120) and negatives (150), while receiver operating characteristic curve analysis confirms high discriminative power (area under curve approximately 0.99). These results indicate the model's efficacy in enhancing IoT security by enabling edge-deployable, low-latency intrusion detection, with potential applications in smart homes and industrial systems for proactive threat mitigation and reliable data flow.