An Edge Computing Gateway Enabled User Behavior Prediction for Personal Smart Home Systems

This paper presents an edge computing gateway system integrating a lightweight hybrid model (Edge1DCLSTM) for user behavior prediction in smart home scenarios. The proposed model combines 1D-CNN and LSTM with only 15,340 trainable parameters, achieving 95.86% classification accuracy on the CASAS Aruba public benchmark, and 92.34% accuracy after local fine-tuning on a real-world home dataset. Deployed on a commercial home gateway, the system implements a full closed intelligent control loop within the local area network, with an end-to-end response time of 50 ms. In high-frequency raw data uploading scenarios, the system reduces daily bandwidth consumption by up to 1600× compared with mainstream cloud-centric solutions. The source code of this work is available at: https://github.com/PandaKing2021/Behavioral-Predictive-AIoT-Gateway.