The suit customization enterprise fails to fully utilized the information about customized style. The FP-growth algorithm in the association rule consumes a large amount of memory with low execution efficiency when it comes to multidimensional big data. Aiming at such issues above, this study proposes an improved mining method for the suit customization based on FP-growth and K-means algorithm. It improves the FP-growth algorithm from three aspects: using hash table to establish item header table, replacing traditional FP-tree with ordered FP-tree, and adding imbalance ratio as the new evaluation index. Experimental results show that compared with other association rule algorithms, the improved FP-growth algorithm reduces the memory consumption by about 6.7% and increases the execution efficiency by about 15%. Through the manual review of experimental results, this algorithm can find meaningful association rules attractive to users, verifying the the proposed algorithm.