A method, an apparatus and a storage medium for stay point recognition and prediction model training are proposed. The method may include: for a to-be-recognized positioning point, obtaining respectively features of N predetermined dimensions for the to-be-recognized positioning point, N being a positive integer greater than one; generating a feature vector for the to-be-recognized positioning point according to the features obtained; determining whether the to-be-recognized positioning point is a stay point based on the feature vector, by using a prediction model obtained by pre-training. The technical solution may be applied to improve the accuracy and a recall rate of the stay point recognition.