To improve the accuracy of sales prediction in a supply chain, an improved forest optimization algorithm(IFOA) is proposed to optimize the prediction of sales volume. Firstly, the delta potential model in the quantum system is introduced to not only make the best of the local optimum information, but also improve the performance of avoiding being caught in local optimum. Then, the adaptive local seeding step is introduced so that both global search and local search velocities are optimized and at the same time, high solution accuracy is ensured. Adaptive transfer rate in the global seeding stage is defined to effectively balance the diversity of forest individuals against the characteristic of local convergence of the algorithm. Furthermore, correlation and the relevant significance between mined exogenous data and the sales data are calculated to carry out feature selection, and features of sales data are extracted based on ensemble empirical mode decomposition(EEMD). Finally, the mentioned features are used to build an IFOA support vector regression(IFOA-SVR) based predictive model so that the model can predict sales well.