Abstract: The basic Golden Jackal Optimization algorithm (GJO) had several drawbacks such as low computation precision, poor exploitation, and ease to get stuck in a local optima when solving high-dimensional optimization problems. An improved GJO algorithm (I-GJO) was proposed. In I-GJO, the original randomly decreasing energy factor was replaced by a nonlinear decreasing factor based on sine function to balance the global exploration and local exploitation abilities of algorithm during the search process. In the later iterative stage of algorithm, a somersault learning strategy was introduced to expand the population search region and improved the solution precision. In order to verify the effectiveness of the proposed I-GJO algorithm, six benchmark function optimization problems were selected for experiment. The experimental results indicated that I-GJO had higher precision and faster convergence speed than the Grey Wolf optimizer (GWO), Seagull Optimization Algorithm (SOA) and the basic GJO algorithm. Finally, I-GJO was applied to solve the feature selection problem. The numerical results on sixteen benchmark datasets showed that I-GJO could effectively remove the redundant features and improve the classification accuracy.