Abstract: The supervised strategy is introduced to improve the granule conditional neighborhood entropy, and an attribute reduction algorithm based on supervised granule conditional neighborhood entropy is proposed. Experiments are carried out on four open datasets. Experimental results show that the proposed algorithm has higher reduction rate and classification accuracy. Based on this algorithm, an analysis model of the fatigue life influencing factors of aluminum alloy welded joints was established. The coupling relationship among the fatigue life influencing factors of the welded joints is analyzed by using the mutual information theory. Analyzing results show that the stress concentration factor is most affected by the joint type and welding method, and least affected by the weld leg length. It indicates that the joint type and welding method should be major considerations when calculating the stress concentration factor. The weight of equivalent structural stress range on fatigue life is 0.461 2, and the weight of nominal stress range is 0.347 3, which indicates that after stress correction, the weight of equivalent structural stress range on fatigue life of the welded joints increases compared with nominal stress range, so it can predict fatigue life of the welded joints more accurately.