The rapid development of Internet of Things technologies has led to the continuous increase and complexity of data feature dimensions. Therefore, a diversity-driven optimization-based adaptive particle swarm optimization feature selection algorithm was proposed to improve the accuracy of feature selection for high-dimensional data. The diversity drive of particles was constructed through population diversity. Secondly, the feature space segmentation method of the algorithm was improved. An adaptive population size adjustment mechanism was proposed. In 12 data sets with different dimensions, the proposed method had an advantage over other methods in terms of average accuracy. The average accuracy of these two classifiers increased by an average of 12.71% and 9.89%, respectively. The average time cost of the proposed method running 30 times on 12 data sets was 343.83ms, which was an average reduction of 44.02% compared to the other three algorithms. Therefore, diversity-driven optimization methods can enhance the algorithmic particle optimization speed. The proposed algorithm requires lower computational costs and superior feature selection accuracy for high-dimensional data feature selection. This algorithm has positive application value in high-dimensional data feature selection problems.