Exploring AI for Smart Homes: Combining Capsule Networks and Bayesian Neural Networks for Enhanced Efficiency

The paper proposes a capsule networks (CapsNets)- and BNNs-abstracted hybrid artificial intelligence (AI) model to improve the performance of smart home systems. It combines the feature-learning ability of CapsNets with the estimation capacity of Bayesian neural networks (BNNs) in uncertainty for smarter decision-making devices in smart homes. The study illustrates an end-to-end approach from data procurement up to preprocessing and, lastly, integration of the two AI models. The hybrid model was tested on a smart home data set with sensor readings (i.e., temperature, movement, appliance usage, and light). Hybrid model performance was compared with basic machine learning (ML) algorithms like XGBoost and random forest. Performance indicates that the hybrid AI model outperforms individual models with 92.3% prediction accuracy, less uncertainty variance, and better energy conservation than the other models. Feature importance analysis highlighted ambient temperature and motion detection as optimal parameters to automate smart homes. Additionally, the hybrid model decreased decision delay by 10% and power efficiency by 21.8% relative to conventional models. Such results rationalize the usability of combining CapsNets and BNNs to create smarter, wiser, more responsive, and more effective smart home systems with the capability to deal with uncertainty and minimize real-time energy consumption.