Resource-constrained environments such as Internet of Things (IoT) and edge networks face increasing cybersecurity threats while lacking sufficient computational capacity for traditional security mechanisms. Existing deep learning-based intrusion detection systems offer high accuracy but suffer from high complexity and lack of interpretability, limiting their deployment in low-power environments. This paper proposes a lightweight Explainable Artificial Intelligence (XAI) framework for secure data transmission that integrates an optimized autoencoder-based anomaly detection model with SHAP-based interpretability. The system is evaluated using UNSW-NB15 and CICIDS2017 datasets under constrained computational settings. Experimental results show that the proposed framework achieves 97.8% accuracy, 96.4% F1-score, and reduces computational overhead by 32% compared to conventional deep learning models. Additionally, the explainability module identifies critical features influencing attack detection, improving trust and transparency. These results are obtained using benchmark datasets commonly adopted in intrusion detection research. The framework provides a scalable, efficient, and interpretable solution suitable for real-time deployment in IoT and edge networks.