Anomaly detection is critical for the reliability and security of dynamic systems, such as the Internet of Things (IoT), which generate large volumes of temporal network data. Existing methods often fail to capture both spatial and temporal dependencies, limiting their ability to detect complex interaction anomalies. To address this, we propose IS-ANED (Interaction Sequential Anomalous Edge Detection), a dual-module framework that integrates Graph Attention Networks (GAT) and attention-based Gated Recurrent Units (GRU) for robust edge anomaly detection. IS-ANED utilizes a hybrid attention mechanism within GAT to prioritize relevant neighboring nodes, minimizing the influence of noisy data. Additionally, an attention-based GRU is introduced to dynamically update node representations, capturing temporal dynamics and overcoming the limitations of traditional GRUs in modeling long-term dependencies. This combination of GAT for spatial propagation and GRU for temporal updates enables IS-ANED to detect anomalies with high accuracy and resilience in evolving networks. Extensive experiments on benchmark datasets demonstrate that IS-ANED outperforms state-of-the-art methods, achieving superior anomaly detection accuracy across both static and dynamic settings, with varying levels of noise and anomaly complexity.