Design of Sensor Data's Missing Value Handling Technique for Pet Healthcare Service based on Graph Attention Networks

Sensor data develops gaps from network and contact failures, and how those gaps are filled changes how the analysis is read. A person can look at a gap and judge for themselves; a companion animal cannot, so a distorted reading becomes a serious problem.

• Uses the fact that pet wearable sensor readings are correlated to derive attention values and feature maps over neighboring nodes
• An LSTM predicts the data, and the missing values are imputed during decoding
• With a variation of the model, the same approach can serve anomaly detection