Smart Home Resource Management based on Multi-Agent System Modeling Combined with SVM Machine Learning for Prediction and Decision-Making

The challenges of the Internet of Things (IoT) in a Smart Home are the monitoring of energy consumption and the automation of the household appliances connected to the Wireless Sensor Networks (WSN). The Smart Home monitoring technology has recently received great attention in the areas of IoT, home automation and WSN monitoring. Many companies seek to address a wide range of important issues including data mining and analysis, energy saving, comfort and security. The Smart Home application is inherently dynamic in the sense that it forms time-series data with real-time changing behavior. We seek to extract and analyze this incoming data to provide and predict useful features for the decision-making system in a Smart Home. This paper describes a new methodology of Smart Home data mining analysis based on Support Vector Machine (SVM) learning for a proposed Multi-Agent System (MAS). The key ideas are to represent the WSN behavior exchange by the modeled MAS, then to predict and classify features using the SVM regression model. Based on the cross-validation performed on the training data-set, the SVM model parameters are optimized for each combination of features. We demonstrate the validity of our methodology in the scenarios of emerging data recognition using a real "Smart Life" database.