Abstract: Objective In order to improve the accuracy and generalization ability of quality prediction for micro resistance spot welding under small sample data conditions, a regression prediction method based on leave one out and particle swarm optimization collaborative optimization kernel extreme learning machine (LOO-PSO-KELM) is proposed. Methods Firstly, the orthogonal experimental method is used to conduct resistance spot welding process experiments, and a small sample dataset is established. The one left method is used for cross validation to classify the dataset. Then, based on the absolute error of the validation dataset and the prediction model of the kernel extreme learning machine, the particle swarm optimization algorithm is used to optimize the parameters of the kernel extreme learning machine, obtaining a reliable and stable prediction model. Finally, based on the selected welding process parameters and the LOO-PSO-KELM model, the particle swarm optimization algorithm is used to optimize the process parameters and obtain the optimal process parameters. Results Compared with traditional PSO-BP neural networks and PSO-KELM algorithms, the LOO-PSO-KELM algorithm performs excellently on various standards, with root mean square errors of 0.0199 and 4.4249 for predicted melt diameter and tensile shear force, respectively. Based on the selected welding process parameters, the LOO-PSO-KELM model is validated. The relative error between the predicted values of the LOO-PSO-KELM model and the experimental verification results is less than 3%. Compared with the optimal parameters under orthogonal experiments, the tensile and shear forces increased by 2%. Conclusion Compared with traditional methods, the LOO-PSO-KELM prediction model has stronger predictive performance. Under small sample datasets, the proposed method demonstrates strong generalization performance and has good application value in micro spot welding of lithium battery connections.