Link Quality Estimation Based On Extremely Fast Decision Tree

The wireless sensor network is a distributed sensor network that is deployed in the monitoring area and a large number of sensor nodes with sense and monitoring environment are constructed by self-organization.Due to the limited resource of sensor nodes and the existence of a large number of interference factors,an effective and real-time link quality estimation model can provide reliable information for the routing protocol to ensure the stability of the network topology,thereby achieving efficient data transmission,improving network throughput,and saving resources improving the life cycle of the network.As an online machine learning method,Extremely Fast Decision Tree(EFDT)has the attributes of simple structure,strong ability to perceive concept drift,and rapid self-update.In different scenarios and at different time periods,by analyzing the relationship between physical layer parameters and packet reception rate,mean value of the average received signal strength indicator,link quality indicator,signal-to-noise ratio are selected as input parameters of the link quality estimation model;The packet reception rate divides the link quality level as the link quality estimation index,converts the link quality estimation problem into a multi-classification problem,and constructs a link quality estimation model based on EFDT;using the Gini factor as the inspiration for the decision node metric;due to high time complexity of the original model,for the research on the structure of the original link quality evaluation model,the EFDT model is optimized by improving the decision node to select the number of samples of the split attribute;the accuracy and time complexity are used to evaluate the model.The paper deploys wireless sensor networks in scenarios with complex surroundings(indoors)and relatively stable environments(corridors and parking lots),and uses an independently developed link quality test platform to collect data.The experimental results show that the improved EFDT model has the best performance;compared with the original EFDT model,fuzzy logic,very fast decision tree model and stochastic gradient descent logistic regression model,no matter the environment conditions are complex(indoor)or relatively stable(Corridor and parking lot),the model we constructed in the paper has higher accuracy and lower time complexity.