The existing productivity prediction model of fractured horizontal well in volcanic gas reservoir has more influence factors, less real samples, and incomplete parameters, therefore, it is difficult to accurately predict the productivity by using conventional methods. In order to quickly and effectively make certain of the productivity of fractured horizontal well in volcanic gas reservoir with existing data, the influence factors are determined by using Grey Relational Analysis(GRA), and the sensitivity of factor weights is considered to amend the algorithm. The improved PSO-LSSVM productivity model is established based on the parameters of Least Squares Support Vector Machines (LSSVM) which are optimized by Particle Swarm Optimization (PSO) algorithm. This model not only makes full use of the characteristics of the LSSVM small samples, which possess the strong learning ability and simple calculation, but also takes the advantages of fast calculation and better global searching ability of PSO. Comparing the PSO-LSSVM model with the BP-LM model, the improved PSO-LSSVM model has less iteration times, higher calculation precision, and more accurate predict results.