Abstract: In order to improve the accuracy of location prediction for mobile users， a method of location prediction for mobile users was proposed based on parallel pattern mining and path matching. Based on the traditional FP-GROWTH algorithm， the method of node load allocation was optimized， and frequent patterns of mobile users were mined on Spark platform. The index-based path similarity algorithm was improved， and the repulsion algorithm based on the shortest path distance was proposed to improve the applicability of missing trajectory data. Experiments on real user trajectory data sets show that the proposed model based on track dissimilarity prediction method has higher prediction accuracy than that of Markov model and Kalman filter model， which is improved by about 7% on average.

Key words: location prediction, Spark, FP-GROWTH, pattern mining, track dissimilarity