Web caching is used to solve the problem of network access delays and network congestion. The intelligent cache replacement strategy directly affects the cache hit rate. This paper proposed a web cache replacement strategy combining greedy dual size frequency (GDSF) algorithm and support vector machine (SVM) re-accessed probability prediction. In the traditional GDSF method, a new objective function is constructed by considering the network object type and object re-accessed probability. The object re-accessed probability is predicted by learning the historical access data through SVM classifier. The simulation results show that compared with the traditional LRU and GDSF schemes, the proposed strategy has a higher request hit rate and byte hit ratio. When the cache size is 16%, the HR and BHR values reached 0.623 and 0.522, respectively.