Spatio-Temporal Adaptive Graph Attention Networks for Sensor Conflict Resolution and Early Fire Risk Prediction (STAGAT-CAFP) Smart Home Fire Detection Benchmark Dataset: Simulated Residential Multi-Sensor Time-Series Data and Synthetic Fire Scenario Metadata (1200 Hours)

Osman Yakubu, Issifu Damba Kanzoni

This repository contains two datasets that were created for Spatio-Temporal Adaptive Graph Attention Network with Conflict-Aware Fire Prediction (STAGAT-CAFP) that is developed for smart home fire prediction and sensor conflict resolution. Together, the datasets provide a benchmark for intelligent machine learning, deep learning, graph neural network, and physics-informed AI models for early fire prediction. In particular, this repository includes a total of 1,200 hours of simulated time series of multi-sensor measurements in various smart homes environments (Simulated_residential_fire_dataset_1200h.csv), as well as a benchmark dataset (stagat_cafp_synthetic_dataset_1200_hours.csv) which contains simulation meta-data, layout of the residences, fire scenarios, sensor configuration, and sensor conflict labels. This data was created based on realistic fire simulations and includes data from normal residential activities, various types of fires, nuisance alarms, and simulated sensor conflicts, which may include information, action, and policy conflicts.  The time series data includes sensor data from various environmental sensors including but not limited to temperature sensors, smoke detectors, carbon monoxide (CO) detectors, humidity sensors, volatile organic compounds (VOC) detectors, etc. The benchmark dataset will include labels that describe fire scenarios, residential layouts, simulation parameters, conflict categories, and annotations.  These datasets can be useful for studies of early fire detection, smart home safety, graph neural networks, spatio-temporal forecasting, sensor fusion, conflict detection, anomaly detection, explainable AI, uncertainty estimation, continual learning, edge intelligence, and physics-informed machine learning.