Abstract: To solve the problem of unreasonable distribution of feature weights in the data-based prediction model, the whale optimization algorithm was included in the population space of cultural algorithm, and a cultural whale optimization algorithm (CWOA) was acquired for optimal distribution of feature weights. First, the root mean square error of the prediction model was used as the fitness function, and the whale optimization algorithm was utilized to iteratively refine the feature weight in the population space. Then, the optimal weight in the population space was placed in the belief space through the defined acceptance function for performance evaluation and double mutation evolution so as to form situational knowledge and normative knowledge. Finally, the influence function was used to update and guide the weights in the population space to further obtain the optimal allocation results of feature weights. By taking the prediction model on the basis of case-based reasoning as an example, University of California Irvine (UCI) standard data set was used to conduct the contract experiments on different distribution techniques for feature weights. Results show that the prediction accuracy of the technique mentioned in this paper after weight distribution is optimal, delivering certain application value in the field of machine learning involving the distribution of feature weights.