Abstract: Aiming at the problem of the received signal strength (RSS) value and dimension change in complex indoor environment, an improved access point (AP) selection method and a random forest (RF) classification algorithm for real-time indoor location are proposed. The improved AP selection method in the off-line phase uses the RSS data variance of the AP and the AP appearance frequency to measure the AP stability and selects the first m stable APs. When the variance is processed, the Laplacian smoothing is performed to avoid the variance of 0, and construct a preliminary fingerprint database. The online phase uses the RF in the integrated learning to vote on the classification result to arrive at the final position. The improved algorithm is compared with the traditional random forest, the improved AP selection fusion weighted K-nearest neighbor algorithm (WKNN) and the information gain (IG)-based AP selection algorithm plus random forest. The experimental results show that the proposed method Compared with the other three algorithms, the positioning error decreased by 29.3%, 23.2%, and 17.2%, respectively, and the positioning time is also improved.