Potassium sorbate is a typical food preservative in daily life. Excessive consumption of the preservative potassium sorbate shall do harm to people’s health seriously. Using orange juice as background solution, 22 sets of samples of potassium sorbate orange juice solution with potassium sorbate content ranging from 0.007 0～0.100 0 g·L-1 were prepared. In this paper, the fluorescence characteristics of potassium sorbate in aqueous solution and in orange juice solution are studied by using FS920 fluorescence spectrometer. Due to the interference of orange juice, the concentration of potassium sorbate no longer satisfies the linear relationship with fluorescence intensity, and the prediction of the concentration of the substance is complicated. In this paper, an improved chicken swarm optimization support vector machine (ICSO-SVM) model is constructed to process the fluorescence spectrum data. Eighteen samples are selected as training set and four samples as prediction set. Under the optimum excitation wavelength λex=375 nm, the fluorescence intensity of each samples in the range of 450～520 nm are taken as input, and the concentrations of potassium sorbate orange juice are taken as output. Firstly, the parameters of the improved chicken swarm algorithm (ICSO) are initialized, then the optimal values of penalty factor C and kernel parameter g of the support vector machine (SVM) are found by training, and the optimal values are input into the ICSO-SVM model. The predicted concentration values of four groups are 0.011 5, 0.026 0, 0.077 0 and 0.092 0 g·L-1, respectively. The mean square error of ICSO-SVM model is 1.01×10-5 g·L-1, and the average recovery is 101.73%. Compared with chicken swarm optimization support vector machine (CSO-SVM), genetic algorithm optimization support vector machine (GA-SVM) and particle swarm optimization support vector machine (PSO-SVM) under the same conditions. The results show that the prediction accuracy of ICSO-SVM model is higher than that of CSO-SVM, GA-SVM and PSO-SVM. Moreover, the improved chicken swarm algorithm is easier to find the global optimal value in the training process and has faster iteration speed. This paper provides a new method for predicting the concentration of substances.