Simulation Analysis of Distance-aware Graph-based Semi-supervised Learning

According to the problem of agglomeration effect of Graph-based Semi-Supervised Learning (G-SSL), this paper studies a Distance-aware Graph-based Semi-supervised Learning (DG-SSL) algorithm, which reduces the agglomeration effect of G-SSL, and holds smaller average estimation error. When compared with the K nearest neighbors (KNN) algorithm, moreover, the DG-SS algorithm can achieve better positioning result by using a small number of labeled samples. Simulation results show that the DG-SSL algorithm effectively resolve the problem of requiring enough labeled samples for Radio Map setup in indoor positioning algorithm. Thus, it reduces the workload and expenditure of establishing the Radio Map.