Research on Decision Tree Based Link Quality Estimation Methods for Wi-Fi Networks

目前机器学习正在成为Wi-Fi链路质量估计的主要工具，模型的泛化能力对于链路质量估计尤为重要。现有方法在训练模型时，需要在不同条件下采集多组数据，建立模型的成本较高，并且多数使用较复杂的模型。提出一种基于决策树的链路质量估计方法，在预处理阶段对数据进行最小值填充和EWMA处理，选择RSSI和RSSI均值作为模型输入参数，利用一组噪声条件下采集的数据建模，分别进行链路质量分类和回归估计。与现有方法比较，提出的方法以较少的数据就可以获得良好的泛化性能。实验结果表明，该方法的平均准确率达到96%，训练时间为0.01 s，回归估计的平均绝对误差为0.025。Currently, machine learning is emerging as a primary tool for Wi-Fi link quality estimation, with the model’s generalization ability being particularly crucial for accurate estimations. Traditional methods often require collecting multiple datasets under varying conditions to train the models, leading to high costs and often relying on complex model architectures. This paper presents a link quality estimation approach based on decision trees. In the preprocessing stage, the data undergoes minimum value filling and exponentially weighted moving average (EWMA) filtering. The model inputs are selected as the RSSI value and its mean, and the modeling is performed using data collected under noisy conditions. This approach performs both link quality classification and regression estimation. Experimental results demonstrate that with a relatively small amount of data, this method achieves excellent generalization performance, yielding a classification accuracy of 96%, a training time of 0.01 seconds, and an average absolute error of 0.025 for regression estimation.