"Scylla serrata IoT Aquaculture Dataset"

"The presented dataset constitutes a rigorously structured, multi-modal, and time-synchronized benchmark for intelligent aquaculture, specifically designed to capture the complex environmental\u2013behavioral dynamics of Scylla serrata within a vertical farming system. It integrates 18,144 high-resolution environmental observations and 12,480 annotated behavioral images collected over 21 days from three vertically arranged tanks under a fixed 5-minute sampling regime, ensuring strong temporal consistency and statistical reliability. The dataset is characterized by a unified representation that combines raw sensor measurements, derived features, behavioral annotations, and system-level metadata, enabling both data-centric modeling and system-aware evaluation. Environmental variables are carefully constrained within biologically validated physiological ranges to preserve realism, while behavioral labels are generated through a hybrid framework incorporating rule-based thresholds, expert validation, and temporal smoothing, thereby ensuring annotation consistency and reproducibility. Furthermore, the inclusion of predefined train\u2013validation\u2013test partitions, imputation flags, and controlled data perturbations (e.g., sensor drift and missing data simulation) enhances its suitability for robust machine learning, anomaly detection, and reliability analysis. Overall, this dataset provides a comprehensive and scalable foundation for advancing IoT-driven, edge-intelligent, and federated learning approaches in smart aquaculture."