Descriptor: WiFi-Granular Activities in Kitchen Dataset (Wi-GRAK)

This paper introduces a comprehensive, multi-modal dataset designed for advancing device-free human activity recognition in kitchen environments. The dataset concurrently captures both WiFi Channel State Information (CSI) and Received Signal Strength Indicator (RSSI) data, enabling analysis at varying levels of detail. It encompasses nine distinct single-person activities, systematically categorized into three fine-grained (e.g., precise utensil movements), three mid-grained (e.g., combining ingredients), and three coarse-grained (e.g., meal preparation) actions. To ensure robustness and diversity, data was collected from three different users, across three separate environmental settings, utilizing a pair of ESP32 microcontrollers as transmitter and receiver. Each unique activity scenario was repeated over 10 trials, resulting in a total of 1,620 structured data entries. This dataset serves as a valuable benchmark for developing and evaluating machine learning models, particularly in exploring the comparative advantages of using CSI and RSSI for granular activity classification in real-world settings.